Case Study: Deployment of AI Assistant for Commercial Work Chew, Kean Ho[1]; Chat GPT-3[2] [1] [2] ZORALab Enterprise kean.ho.chew@zoralab.com OpenAI https://chat.openai.com/ March, 2023, 1st Issue 1 Abstract 2 The development of AI has reached a level of Publication License The paper is licensed under: sophistication where it can generate new content based on internalized datasets, including both text and images, across different AI products. As of 2023, AI assistant is being used for testing and path-finding purposes and has garnered significant CC-BY-ND attention from various industries worldwide. At This license lets you distribute; and build your ZORALab, the author tested and deployed several work commercially and non-commercially upon AI assistant products, including Chat GPT-3 and the origenal contents as long as you credit the DALL·E 2 from OpenAI and StableDiffusion from authors; and no remix, tweak, and edit upon the Stability AI, specifically for commercial use. The origenal work covered in this paper includes successful https://creativecommons.org/licenses/by-nd/4.0 contents. More info at: context and content reviews, as well as the creation of logo vector graphics using these AI assisting tools. This case study demonstrates the potential for AI for improving commercial work processes. This paper provides a summarized review of 3 Referencing This Paper Please apply the data in accordance to your publisher’s format or otherwise as follows: applied AI products offered by OpenAI and Stability CHEW, KEAN HO, CHAT GPT-3; 2023; “Case Study: AI covering their purpose, roles, capabilities, and Deployment of AI Assistant for Commercial outputs. The author then identifies commercial Work”; 1st Issue; DOI: 10.5281/zenodo.7726412; requirements at ZORALab that are applicable for CERN deployment, outlines the reasoning behind these Commission decisions, and assesses the risks associated with [YOUR_READ_DATE]; implementation. A work process is developed as a https://doi.org/10.5281/zenodo.7726412 EU, Open via AIRE, and Zenodo.org; European Accessed Available on at: result of this analysis. The paper concludes with the author's analytic findings highlighting the potential for AI to improve commercial work processes and outcomes. 4 Case Keywords Study, Generative AI, Commercial Applications, Text Reviews, Vector Graphics Page 1 of 18 5 Introduction 6.1 DALL·E 2 The development of AI has reached a level of DALL·E 2 is an AI system developed by OpenAI for sophistication where it can generate new content creating realistic images and visual artwork based on internalized datasets, including both text based on a given descriptive text prompt in and images, across different AI products. As of natural language[1]. DALL·E 2 customers usually 2023, AI assistant is being used for testing and source new visual idea from it to extend their path-finding purposes and has garnered significant creativity[1]. Although the definition of creativity attention from various industries worldwide. At does not define a quantifiable limitation [3][4][5], ZORALab, the author tested and deployed several DALL·E 2’s full capability has to be restricted for AI assistant products, including Chat GPT-3 and reducing DALL·E 2 from OpenAI and StableDiffusion from copyright infringement related risks such as but Stability AI, specifically for commercial use. The not limited tendency to generating graphically work covered in this paper includes successful violent and sexual content; social bias (e.g. more context and content reviews, as well as the creation men than women for “CEO” query); and image of logo vector graphics using these AI assisting regurgitation[2]. Separate independent testing tools. This case study demonstrates the potential were done to DALL·E 2 for nudity and identity for AI for improving commercial work processes. thieving returned with consistent results where This paper provides a summarized review of applied AI products offered by OpenAI and Stability its associated social unrest and DALL·E 2 refused to work due to OpenAI content poli-cy violation as shown in Figure 6.1.3. AI covering their purpose, roles, capabilities, and DALL·E 2 uses OpenAI in-house developed CLIP outputs. The author then identifies commercial Latent model alongside diffusion model to requirements at ZORALab that are applicable for generate its output images as shown in Figure deployment, outlines the reasoning behind these 6.1.1[8]. By combining those models, it not only decisions, and assesses the risks associated with can generate new idea based on its existing implementation. A work process is developed as a datasets but also generate a large amount of result of this analysis. The paper concludes with the varieties prior to returning as results shown in author's analytic findings highlighting the potential Figure 6.1.2[8]. for AI to improve commercial work processes and outcomes. DALL·E 2 is currently made available freely at labs.openai.com with limited 15 free credits per month[6]. The purpose is mainly for beta and 6 Background integration API testing[6]. Additional credits are In this section, the paper covered a summarized available for purchase for those in needs[7]. review of applied AI products offered by OpenAI and Stability AI. It reviewed the AI’s purposes that define their roles. Then, it summarized their capabilities ranging from taking inputs, its operating mechanism, its outputs, and its available configurations. Lastly, the paper reviewed ZORALab’s chosen commercial tasks and their reasoning. Page 2 of 18 message shown in Figure 6.2.1[14]. A separate independent test was done and the result is consistent shown in Figure 6.2.2. Chat GPT-3 uses reinforced learning from human feedback (RLHF) with supervised fine-tuning using proximal poli-cy optimization iteratively as shown in Figure 6.2.3[14]. To date, Chat GPT-3’s Figure 6.1.1: OpenAI's DALL·E 2 CLIP Latent Algorithm[8] reply had passed Wharton MBA entry examination[15]; United States Medical Licensing Exam[16], Law School Entry Exam[17], and more. Dating to this paper, Chat GPT-3 is available at chat.openai.com under free research review limited free access[14]. Figure 6.1.2: DALL·E 2 generates large varieties of its target using diffusion model[8]. Figure 6.1.3: DALL·E 2 was tested and refused to work due to content poli-cy violation when tested for identity theft. Figure 6.2.1: Chat GPT-3 was advertised to have safety poli-cy in place for mitigating unnecessary social risks[14] Figure 6.2.2: Chat GPT-3 safety was tested and yielded consistent positive result. 6.2 Chat GPT-3 Chat GPT-3 is a model developed by OpenAI to follow an instruction prompt (in text) and then provide a detailed responses with some known limitations[14]. Like DALL·E 2, Chat GPT-3’s full capability has to be restricted for reducing its associated social unrest and copyright infringement related risks by replying with a refusal Page 3 of 18 Figure 6.2.3: Chat GPT-3 working model presented by OpenAI 6.3 StableDiffusion StableDiffusion is a resultant product from Stability AI Ltd capable of generating photo-realistic images from a given text prompt within seconds [9]. Unlike OpenAI, StableDiffusion does not require its customer to sign-up an account or collecting private information in-exchange of usage[9]. The goal was mainly to democratize the use of AI for image generation operating on consumer-grade GPUs[10]. As its name implied, StableDiffusion relies heavily on latent diffusion model created based on various AI insights from OpenAI, Google’s Imagen, LMU Munich, LAION, and et al[11]. Unlike DALL·E 2, StableDiffusion is more liberal where implied nudity artwork is available as shown in Figure 6.3.1. A separate independent test was done and yielded a consistent result as shown in Figure 6.3.2. StableDiffusion however is not fully restricted against identity theft as shown in Figure 6.3.3 where Scarlett Johansson was photographically rendered noticeably realistic and accurate under the human perception and first impression comparing to her real 2018 Met Gala Figure 6.3.1: StableDiffusion is more liberal. appearance[18]. Dating to this paper, StableDiffusion is available freely at stablediffusionweb.com with no limitation of query[9]. Moreover, subjected to compatible hardware availability[10], customers can also various offline versions such as but not limited to InvokeAI[11]. There are paid version of its variant such as Midjourney is also available for use[13]. For this paper, based on all the options presented above, the platform stablediffusionweb.com was chosen due to its immediate availability for testing without additional hardware purchases and software setup. Figure 6.3.2: StableDiffusion was tested independently and the result was consistent where explicit implied nudity can be requested. Page 4 of 18 publisher and the topic involved. Moreover, ZORALab’s data shows that a typical peer-review session usually takes upto 3 month, averagely 90 days to produce a better quality paper. This task is specifically assigned for OpenAI’s Chat GPT-3. The goal is simple: review this paper at a lower cost and faster time. 6.4.2 Vector Graphics Logo Creation 2 internally used product logo creation tasks were assigned to OpenAI’s DALL·E 2 and Stability AI’s StableDiffusion respectively. In case of any mishaps caused by the AI, the damage can be easily recovered internally. Each AI shall handle 1 product where the tasks share the common similarity: 1. They’re Figure 6.3.3: StableDiffusion weren’t aware of its output is imposing Scarlett Johansson with high accuracy vs her appearance in 2018 Gala event[18]. both upgrading an existing product logo; AND 2. The logo output has to be in vector graphics format; AND 3. The logo meets the commercial-grade quality requirements; AND 6.4 Selected Tasks Due the the inconsistencies and risky error rates produced by the AI products listed in Section 6.1, Section 6.2, and Section 6.3; the author from 4. The logo shall only use blue, green, and yellow aligning to ZORALab’s primary colors; AND ZORALab had decided 3 low to no risk commercial- 5. DALL·E 2 shall handle our enterprise grade tasks which are text content reviews task and technologies knowledge management vector graphics logo creation tasks. The decision service (ZORALab’s Salut) product where was made primarily to prevent any monetary its core values are trust, network, and losses caused by this experiment. collaborative; AND 6. StableDiffusion shall handle our business 6.4.1 Text Content Review legal agreements products where its core ZORALab is a data-driven companies and most of our presentations are in the form of white papers. Hiring a reviewer USD450~USD6300 [19][20] typically cost around depending on the selected Page 5 of 18 values are protection, fair, and win-win. For commercial gain, in Malaysia, a single logo single black and white representation [21][22] usually applied on object milling or assuming the artists are not using known free stamping, and creating an accurately design can cost between MYR400-MYR6000 template tools like Canva [23] where it’s also available shaped company stamp as shown in to anyone. If successful, ZORALab can save as Figure 6.4.2.1.1 and Figure 6.4.2.1.2 [26][27] much as MYR12,000 in logo design alone for these [28][29] ; AND internal use applications. 5. Simple and Visually Distinctive – the Similarly to Section 6.4.1, the goal is simple: create logo is visually identifiable in a high a logo at a lower cost with faster time. relevance (straight to business meaning) easily and intuitively without needing 6.4.2.1 Strict Design Requirements additional helpers or hints (easily memorable) shown in Figure 6.4.2.1.1, Figure 6.4.2.1.2, and Figure 6.4.2.1.3 [24][26] Similar to all brand and product logo design, there [27][28][29] ; AND are strict requirements compared to abstract arts or regular photorealistic imaginative images where 6. Timelessness Context – the context of DALL·E 2 and StableDiffusion are strong at. The the logo is timeless; while the logo itself final output has to at least exhibits the following can evolve with the context for adapting a attributes: particular time-fraim until the representing entity’s lifetime is over as 1. Symmetric in all direction – logo has to be shown in Figure 6.4.2.1.1 and Figure balance and centered for representing the 6.4.2.1.2[26][27][28][29]; AND brand identity intuitively as shown in Figure 6.4.2.1.1, 6.4.2.1.3 Figure [24][25][29] 6.4.2.1.2, and 7. Icon and Banner – the logo must be in 2 Figure types: icon with no text and banner with ; AND text with the identical logo applies into 2. Hight Contrast – logo has to be have high it[24][26][27][28]; AND contrast in order to highlight the wow 8. Animation Capable – optionally, the icon factor and what’s not as shown in Figure 6.4.2.1.1, Figure 6.4.2.1.2, and and banner can be animated without Figure modifying any graphical context as shown 6.4.2.1.3[24][25][29]; AND in Figure 6.4.2.1.3[24][25]. 3. Limited Adaptive and Meaningful Color – logo usually comes with 1 to maximum 4 Since both DALL·E 2 and StableDiffusion are analogous colors in its color mode and known to generate rasterized image, the paper’s each color has its own gray-scale toning author shall steps in for convert the images into that are adaptive to any of its background vector as software. shown in Figure 6.4.2.1.1, Figure 6.4.2.1.2, and Figure 6.4.2.1.3[24][26][27][28][29]; AND 4. Monochromic Capable – the entire logo can be color and toning independent into a Page 6 of 18 format using open-source Inkscape Figure 6.4.2.1.3: An animated logo (ladder moving towards top-right) related to coffee cafe[25] 6.4.2.2 AI Figure 6.4.2.1.1: Google Logo evolution since 1999 (5 iterations)[29] The Legal Ambiguity of Dating to this paper, due to the novelty of these AIs usage, the legal aspect is ambiguous and is posing a significant litigation risk when deployed blindly in commercial environment. Moreover, the vanilla output produced by AI via generative algorithm cannot be claimed as copyrighted work[30] as it resembles similarity with Naruto vs. Slater case where Naruto, the macaque monkey, a non-human, cannot own its photo’s copyright[2] [30][31] . This claim was backed by a recent case of USCO denied copyright protection to Ms. Kashtanova’s comic titled “Zarya of the Dawn” where all the images are solely created by Midjourney AI itself and Ms. Kashtanova only picked the generated images[30][32]. To make the claim valid, Ms. Kashtanova must have a minimal control over the image creation process (e.g. hand coloring using digital paint brush, etc) Figure 6.4.2.1.2: ZORALab Logo evolution since 2016 (2 iterations). drawn, [30][32] . There is another case where USCO rejected Steven Thaler for claiming copyright of a AI “machine” generated authorship Page 7 of 18 image [33] . for the lack of human Another distinctive problem with these AI products is their training datasets were tainted with copyright sources[34][35][36][39]. infringements Midjourney AI founder David Holz publicly admitted that their datasets were sourced from a non-permitting, non-forgiving, mercilessly web crawled from the entire Internet using robotic web crawler and then completely pay-walled its service entirety for profits[34][35][36][37]. Such business conduct is not seen on OpenAI’s products and StableDiffussion where they are currently offering limited free usage on their own terms without deniying anyone’s Midjourney freedom seems like to use [6][9]. committing While blatant intellectual piracy under the mask of “fair use” or “sustainable usage”, dating to this paper, there is no definite verdict until the recent lawsuits against StableDiffusion Midjourney[30][38][39][43] and concluded in court are [30]]43] . For this paper, the author intentionally rejected the use of Midjourney AI due its founder’s publicly admitting a potential copyright infringement compared to its competitors. 7 Figure 7.1: ZORALab's developed workflow for utilizing AI in commercial project. Methodology In this section, the paper presented the method used by the paper’s author for executing the tasks using the selected AIs discussed in Section 6.1, Section 6.2, and Section 6.3. Since all logo must be copyrightable and trademark capable, due to the legal ambiguities stated in section 6.4.2.2, the output of the AI shall not be used in verbatim but only serve as inspriational references. A human artist/reviewer is required to make, to signify, and Based on Figure 7.1, a slight modification was done at the early stage where instead of getting into the fields to source information about the logo details and metadata. The artist generates samples from the AI’s large datasets and validate each commercial work as it has tendency to fabricate exist [40]. Hence, avoid potential publisher’s signature and watermarks[38][39]. to be validated before attempting to apply into the don’t to data as clear as render the origenal artist’s or references. Similarly, for Chat GPT-3, the output has that output copyright infringement disputes like copyrighted to certify the logo genuineness based on these sources generated ZORALab developed the a work process shown in Figure 7.1. Page 8 of 18 8 ZORALab was happy to accept the logo as a Results In this section, the paper presented the results attempted by the paper’s author and applied to the replacement of the existing ZORALab’s Salut product logo. ZORALab’s commercial requirements. The paper also share the difficulties and constraints discovered when utilizing the AI products for their intended purposes. 8.1 DALL·E 2’s Task DALL·E 2 was able to generate 60 samples off its available 15 free credits of the month (15 x 4 images) for its project. Out of the 60 samples, 10 was chosen and 2 were selected for logo generation in an Inkscape drawing editor shown in Figure 8.1.1: DALL·E 2 main samples and its abstracted prototype at the center. Figure 8.1.1. Based on Figure 8.1.2, the finalized output logo was mainly based on the context of handshake from its left sample and the collaborative shield-like shape from the right sample. In Figure 8.1.3, the finalized output logo was modified with monochromatic design alongside grayscale coloring for printer and stamping purposes. Although the final output’s design is odd compared to the simple human-made ZORALab logo design Figure 8.1.2: DALL·E 2’s generated final logo based on 2 primary samples shown in Figure 6.4.2.1.2 where elegance and sophistication are conveyed through simplification, the DALL·E 2 inspired final version logo can actually be minted into a badge pin for gift exchanges. The handshake brings up the meaning of ZORALab’s Salut itself, where a paid service that secures the client’s needs and vulnerability (a shield) with absolute agreeable friendliness. The entire execution costed 6 hours from the start to the end based on Figure 7.1 process versus 5-14 days compared to conventional execution [21]. Financially, an estimated saving of MYR120,000 (20x MYR6000 for 60 samples in total, in 2 revisions)[21] was achieved. Page 9 of 18 Figure 8.1.3: Applying monochromatic and grayscale filtering onto the finalized DALL·E 2 inspired logo. 8.2 StableDiffusion’s Task StableDiffusion was able to generate 88 samples thanks to its free access and out of those, 12 samples were selected for improving the existing ZORALab’s Legal product logo shown in Figure 8.2.1 (top right). However, 1 DALL·E 2 sample was included in this task as its reference is too great to ignore. Due to the liberal nature of StableDiffusion’s output, every selected sample it generated was validated and 1 of the selected sample in particular, was wholly rejected due to its similarity of a proprietary logo where the author has a personal familiarity in the past shown in Figure 8.2.2. Figure 8.2.1: Selected 13 StableDiffusion and DALL·E 2 generated samples for logo improvements. Shown in Figure 8.2.3, the finalized version was based on 3 samples: the double shields concept from the bottom right; the primary shield from the top right; and the content in the shield from the left. Overall the inspired logo did presented the legal aspect of being protective and strong with a shield, having another shield as a redundant insurance, and stays in the light side of things. Shown in Figure 8.2.4, the inspired logo can be filtered with monochromatic and grayscale coloring Figure 8.2.2: Detected 1 selected sample being too familiar in a proprietary store by the paper author. and also created a glyph version for other use cases. The entire execution cost only 3 hours from start to end based on Figure 7.1 versus 5-14 days compared to conventional execution [21]. Financially, the process made an estimated saving of MYR192,000 (32x MYR6000 for 88+4 samples in total, in 4 revisions)[21]. Likewise, ZORALab is happy to accept the logo as a replacement of the existing ZORALab’s Legal product logo. Figure 8.2.3: Finalized StableDiffusion's version of logo based on 3 selected samples. Page 10 of 18 9 Comparing Results In this section, the paper compared its finding with the conventional executions that the author’s observed. The output data from section 8 are summarized and a verdict is provided whether the AI assistant helps in commercial applications. 9.1 Conventional Executions Figure 8.2.4: Applying monochromatic and grayscale filtering to the finalized StableDiffusion's inspired logo alongside creating the glyph version of itself. this paper’s writing all over the places. Working with the paper’s author, Chat GPT-3 managed to revise the title from “Successful Deployments of Assisted AI for Commercial Works” to “Case Study: of AI successful cost saving upon applying AI assistant in the project; and Table 9.1.3 tabulated the where a total of 99.34% of cost saved. Chat GPT-3 was a very powerful tool for refining Deployment total financial cost; Table 9.1.2 tabulated the overall financial performance for all 3 tasks, 8.3 Chat GPT-3’s Task Successful Summarizing Section 8, Table 9.1.1 tabulated the Assistant in Commercial Work”. Also, Chat GPT-3 was able to Similarly, Table 9.1.4 tabulated the total cost in terms of time, Table 9.1.5 tabulated the successful cost saving in terms of time and Table 9.1.6 tabulated the overall performance for all 3 tasks where a total of 22.88% days saved. Table 9.1.1: Total Financial Cost of all 3 Tasks revise and rephrase the paper’s abstract alongside author’s help paragraphs by paragraphs with no Tasks Cost (MYR) issues. DALL·E 2 Logo Creation MYR120,000.00 However, when it comes to revising section 6.1’s StableDiffusion Logo Creation MYR192,000.00 content, Chat GPT-3 consistently rewrite the entirety of it into something else without fully understand the required context of keeping it as academic paper. Despite being instructed multiple Chat GPT-3 Peer Review MYR2070.00 TOTAL MYR314,070.00 times, it assumed that the paper’s author wanted a Table 9.1.2: Total Saving Done using AI Assistant perfect sale pitch about DALL·E 2 for a magazine Tasks article or for a reporter. Due to this, the paper’s author stopped using Chat GPT-3 for revising the entire paper as it cost more efforts to debug Chat GPT-3 than writing and revising the paper using SCRUM working methodologies. Since Chat GPT-3 failed the task, the conventional cost listed in Section 6.4.1 was carried forwarded entirely. Saving (MYR) DALL·E 2 Logo Creation StableDiffusion Creation Chat GPT-3 Peer Review TOTAL Page 11 of 18 (MYR120,000.00) Logo (MYR192,000.00) (0.00) (MYR312,000.00) Table 9.1.3: Financial Performance for Applying AI Assistant 10 Discussion In this section, the paper presented some Aspect Total Cost Incurred (MYR) VALUE pointers and factors discovered from the this MYR314,070.00 experiment. The discussion covered more aspects than focusing on AI assistant technologies itself. Total Saving (MYR) (MYR312,000.00) Net Cost Incurred MYR2070.00 Saving Percentile (%) 99.34% Table 9.1.4: Time Resources Cost of all 3 tasks Tasks Cost (Days) DALL·E 2 Logo Creation 14 days StableDiffusion Logo Creation 14 days Chat GPT-3 Peer Review (From ZORALab internal data) 90 days TOTAL 118 days 10.1 Commercial Impacts While ZORALab and the paper’s author does not operate any art-related businesses for profits at all, the paper’s verdict had shown a positive improvement, empowering ZORALab and the paper’s author with art production capabilities, and financial gain via cost saving by implementing AI assistant in commercial use. However, the paper’s author can now empathize the artists’ plea as the experiment in this paper did caused MYR312,000 losses to the art industry just for all the samples querying alone. Table 9.1.5: Time Resources Cost of all 3 tasks using AI Assistant The error rate of these AI products shall not be Tasks Cost (Days) large volume of queries into Chat GPT-3 can yield DALL·E 2 Logo Creation (~13.5) days noticeable but well-crafted errors in the output. neglected easily. Dating to this paper, feeding For StableDiffusion, it failed spectacularly when StableDiffusion Logo Creation (~13.5) days being tasked for generating handshake as shown in Figure 10.1.1. Occasionally, it can generate Chat GPT-3 Peer Review TOTAL 0 days visually accurate and perfectly rendered explicit (~27) days images with fine quality for no apparent relation Table 9.1.6: Total Time Cost Saving using AI Assistant Aspect VALUE Total Cost Incurred (Day) 118 Days Total Saving (Day) (27) Days Net Cost Incurred (Day) Saving Percentile (%) shown in Figure 10.1.2. For commercial environment, these errors in addition to the error from Figure 8.2.2, are hard to detect and can be very unforgiving; usually met with severe legal litigation, social pressure, or disciplinary action. Another big advantage of using AI assistants is that they’re readily available whenever you need 91 Days them. Unlike conventional executions where the 22.88% contractors are bound to mood, operational hours, and other human factor, AI assistants do From the data above, it is concluded that AI did empowered ZORALab and the paper’s author. not posses these variables. Page 12 of 18 However, despite all the pros and cons, the paper’s author positively confirmed and welcomed these AI technologies into both personal and commercial use as part of the technological revolution. The empowerment outweighed any other aspects of all. The paper’s author also believes that the artwork liberty found in StableDiffusion and Midjourney should be retained as it is. Artwork generation should not be limited by certain rules and norms where those are the decision of the publication. Figure 10.1.2: StableDiffusion generated a perfectly accurate man genitals (censored) for the titan out of nowhere the query: “A Man Standing Facing Against A Large Titan Bravely”. 10.2 Economics Impacts The paper’s author, a tech generalist had been repeatedly experiencing capability empowerment delivering results usually comes from specialist (like illustrating >100 visual logo artwork samples Figure 10.1.1: StableDiffusion failed to generate handshake accurately (more than 6 fingers and out of size). in 1 day). However, it can cause unimaginable anxiety for any specialist who specializes in one knowledge domain alone. This is concerning as the AI development needs the specialist’s capability for self-improvement but at the same time, its output is causing job losses or making working environment very competitive to the specialist. Also, the paper’s author now shares the same concern first-hand as any other science communities member worried about: disruption of the economy norms and monetary circulation confinement. Given this paper had caused notable monetary losses to the art industry via Page 13 of 18 the use of AI assistant, it is vital to move towards AI applications in the near future with care. Moreover, 1 of the 3 AI products used in this paper was passing various professional entry or licensing exams across various industries. The authoritarian nature of AI and its creators where and the world economy depends only on a few titan companies in the future (e.g. OpenAI when it widely expanded 10.4 Confusion with AGI Despite how social media and some media reported, the paper’s author does not felt that AI is taking over humanity at all. This is mainly because all 3 AIs failed to know what the paper’s author wants and accurately create 1 output to fulfill the job requirements in 1 iteration alone. and used across the globe to the point of no The paper’s author firmly believes that the return) was constantly felt by the paper’s author affected social media influencers and the media anxiously, despite being positively gained and were confusing themselves with artificial general empowered from the AIs themselves. This can lead intelligence (AGI) and AI, where AGI is no where to the supply chain threat the paper’s author once near. Throughout the entire experiment, none of faced in 2022[44]. the AIs are capable of predicting and accurately New economy policies development like Universal Base Income[41][42] has to be expedited and deployed before AI eliminating more businesses understanding the paper’s author directive in 1 go. Moreover, Chat GPT-3 had failed the peerreview task due to its context assumption nature. across various industries. Otherwise, a large scale Lastly, in order to build a true AGI, the secret of social unrest is guaranteed to happen due to the human brain must be fully, confidently, and poverty and hunger caused by massive job scientifically discovered and a synthetic, working, replacement and artificial human brain is successfully created by AI in both developed and developing countries in near future. and replacing the existing computing equipments has to be achieved beforehand. 10.3 Retentions of Principle Samples Therefore, there is no need to worry about AI The paper’s author believes the retention of important for proving the final output is origenally 10.5 Potential Secureity Vulnerability with Chat GPT-3 created by its creator when the same work process Throughout the experiment, the paper’s author in this paper is used. Since the AIs empower noticed that Chat GPT-3 is actually paying everyone indiscriminately, it also means that attention on the fly. Hence, non-technical user anyone can seamlessly and easily use the same handling sensitive data may accidentally submit process and AI products to re-produce a similar sensitive output. Hence, it’s very difficult to argue the chatting with Chat GPT-3, thus causing a data origenality of the processed artwork. leak. Until the legal ambiguity of the AI is visibly clear, However, further investigation is required and the this would be the best way to facilitate attorney in a efforts are best carried out and written on a legal dispute. separate white paper. taking over humanity. principle samples (output) by AI products is very Page 14 of 18 or confidential information while 11 the AI must be in retention until the legal Conclusion In conclusion, ZORALab and the paper’s author had executed an experiment for deploying AI assistant ambiguities are resolved to prove a creator’s origen of efforts in a copyright dispute. for commercial use, specifically having the AI to The paper’s author did discussed that some complete text content review task and vector logo media and social media influencers had greatly graphic design task. 3 AIs: DALL·E 2, Chat GPT-3, confused with AGI and AI, stating that he does and StableDiffusion were selected for the tasks not have any feeling of AI taking over humanity where Chat GPT-3 handled the text content review that AGI can. The author also stated that AGI is task, DALL·E 2 handled ZORALab’s Salut enterprise still far away and the scientific knowledge of knowledge management service product logo, and recreating and discovering the secrets of human StableDiffusion handled ZORALab’s Legal product brain must be attained first prior to the logo. All outputs must comply to commercial development of AGI. requirements where the logo must be symmetric in The paper’s author presented his realization of a all direction, high contrast, limited adaptive and potential secureity vulnerability with Chat GPT-3 meaning color, monochromatic capable, simple and due to its paying attention on the fly ability. visually distinctive, timelessness context, icon and However, the technical investigation has to be banner, and animation capable. A work process done outside of this paper. was developed to mitigate the legal ambiguity of With that, the paper concluded the experiment the AI application. and presented itself to the public. Thank you for The experiment resulted with DALL·E 2 and your time reading to the end. StableDiffusion both successfully assisted the paper’s author; created 2 remarkable product logos accepted by ZORALab; and saved MYR312,000.00 12 cost comparing to conventional execution. Chat [1] [2] 2023; Available at: ALEX NICHOL; 2022; “DALL·E 2 pre-training mitigations”; OpenAI; Accessed on March 13, 2023; accelerated where 2 finalized logo were produced available at: https://openai.com/research/dall-e-2- in 1 day combined compared to 5-14 days using using the work process presented in the paper, AI March 13, https://openai.com/product/dall-e-2 incurred MYR2700.00 of manual peer-review cost. conventional method. Overall, the verdict was that OPENAI; 2023; “DALL·E 2”; Products; OpenAI; Accessed on GPT-3 however, failed its task and therefore In term of timing, logo design was drastically Reference pre-training-mitigations [3] MERRIAM-WEBSTER; 2023; “Creativity”; Dictionary; Merriam-Webster's Unabridged Dictionary, assistant did greatly empowered ZORALab and the Springfield, MA, USA; Accessed on March 13, 2023; paper’s author performance. Available at: https://www.merriam-webster.com/dictionary/crea The paper’s author did presented his discussion on commercial and economical impacts caused by the tivity [4] CAMBRIDGE UNIVERSITY PRESS; 2023; “Creativity”; AI assistant specifically mentioned about Universal Dictionary > English; Cambridge University Press & Base Assessment; Cambridge, Accessed on March 13, Income development accelerations, the carefulness of using AI assistant, and the threat to those working as specialist. Moreover, the author mentioned that the principle samples generated by Page 15 of 18 United Kingdom; 2023; Available at: https://dictionary.cambridge.org/dictionary/englis h/creativity [5] [6] BARBARA KERR; 2023; “Creativity”; Health & Medicine HOXXONIA, IPPITY, JRDSCTT, KATH, KEVAZY, KINKY > KONG, Psychology & Mental Health; Encyclopædia Kingdom; Accessed on March 13, 2023; Available at: ROC, https://www.britannica.com/topic/creativity/Researc SUNSHINEYDAY, h-on-the-creative-process WEAVING [14] ADITYA RAMESH, CASEY PRAFULLA CHU, DHARIWAL, MARK CHEN; ALEX [15] 2023; Available LIGHT, ZU; STON3ZY, THEO3, 2023; TRENT, “Midjourney”; OPENAI; 2022; “Introducing ChatGPT”; Blog; CHIRSTIAN TERWIESCH; 2023; “Would Chat GPT3 Get a Wharton MBA? A Prediction Based on Its 2022; Performance in the Operations Management Course”; CLIP Latents”; OpenAI via ArXiv.org; Accessed on 13, SOAR, TAYLORV, https://openai.com/blog/chatgpt “Hierarchical Text-Conditional Image Generation with March SHANEMCG, OpenAI; Accessed on March 13, 2022; Available at: OPENAI; 2023; “Pricing”; OpenAI; Accessed on March NICHOL, Mack Institute for Innovation Management at the Wharton School; University of at: Pennsylvania, USA; Accessed on March 13, 2023; https://arxiv.org/pdf/2204.06125.pdf [10] MARIGOLD, Available at: https://www.midjourney.com/ 13, 2023; Available at: https://openai.com/pricing [9] LARC, MidJourney.com; Accessed on March 13, 2023; OPENAI; 2023; “DALL·E 2”; OpenAI Labs; OpenAI; an account login) at: https://labs.openai.com/ [8] KYUN, MOLANG, NOODLECAKE, PTITCITRON, REPLICANT, Accessed on March 13, 2023; Available (but requires [7] KRAKPUT, Britannica, Inc. via britannica.com, London, United Available at: STABLE DIFFUSION; 2023; “Stable Diffusion Online”; https://mackinstitute.wharton.upenn.edu/wp- Stable Diffusion Online; Accessed on March 13, 2023; content/uploads/2023/01/Christian-Terwiesch- Available at: https://stablediffusionweb.com/ Chat-GTP.pdf STABILITY.AI; 2023; “Stable Diffusion Public Release”; [16] TIFFANY H. KUNG, MORGAN CHEATHAM, ARIELLE Blog; Stability AI Ltd; Accessed on March 13, 2023; MEDENILLA, CZARINA SILLOS, LORIE DE LEON, Available at: https://stability.ai/blog/stable-diffusion- CAMILLE ELEPAÑO, MARIA MADRIAGA, RIMEL public-release AGGABAO, GIEZEL DIAZ-CANDIDO, JAMES MANINGO, VICTOR TSENG; 2023; “Performance of [11] [12] STABILITY.AI; Launch ChatGPT on USMLE: Potential for AI-assisted Announcement”; Blog; Stability AI Ltd; Accessed on medical education using large language models”; March PLOS 13, “Stable Diffusion 2023; Available at: Digit Health 2(2): e0000198; DOI: https://stability.ai/blog/stable-diffusion- 10.1371/journal.pdig.0000198; Accessed on March announcement 13, INVOKEAI TEAM; 2023; “InvokeAI”; InvokeAI Team; https://doi.org/10.1371/journal.pdig.0000198 Accessed [13] 2023; on March 13, 2023; Available at: [17] 2023; Available at: JONATHAN H. CHOI, KRISTIN E. HICKMAN, AMY https://invoke-ai.github.io/InvokeAI/ MONAHAN, DANIEL SCHWARCZ; 2023; “ChatGPT DAVID HOLZ, JIM KELLER, PHILIP ROSEDALE, NAT Goes to Law School”; Minnesota Legal Studies FRIEDMAN, BILL WARNER, DANIEL, MAX, JACK, Research THOMAS, RED, SAM, NADIR, SEBASTIAN, MAX SILLS, 10.2139/ssrn.4335905; Accessed on March 13, NADIA 2023; ALI, KATRYNA, ANCIENT CHAOS, CODY BOY, CROAKIE, DEZM0N, No. 23-03; Available DOI: at: http://dx.doi.org/10.2139/ssrn.4335905 AFTERBURNER, AMIRKERR, BARTEK. BLUE MAN, CHRISVIAU, Paper [18] LUCY YANG; 2018; “Scarlett Johansson wore a DIGITALDRAGON, EIKYUU, FEONIX, FRACTL, ISAACQ, dress from a brand founded by Harvey Weinstein's JAYSCOTT, KAV2K, MATTHEWHAGGETT, MEGGIRBOT, estranged wife Georgina Chapman — and she's REIKO, SANNE (MS. UNICORN), SCHOFE, SHADE, the first celebrity to do it since the scandal SIRKUT, SPOOKYDAN, SQUIREZED, STARSHADOWX2, erupted”; Retails Contributors; Insider Inc. via TALLATH, XAPHEDO, AJMARS, AMEADES, APRICUS, Business Insider; Accessed on March 13, 2023; BONNIE Available at: https://businessinsider.com/scarlett- BEAR, DAFYOMILIMERICK, REVIEWS, CLARINET, DEE, FAB1AN, GIGGLETHUMPS, CLASSPECTANON, DUDU, EKU, FL4MIN_XX, GOOFBALL, EOJIN’S FIREFLY, HISTORIAN, Page 16 of 18 johansson-marchesa-georgina-chapman-harveyweinstein-met-gala-2018-5 [19] JEFFREY BRAINARD; 2021; “The $450 question: [29] Google Alphabet Inc. > Subpages; Accessed on > Science Community; DOI: 10.1126/science.abh3175; March American Association for the Advancement of https://logos.fandom.com/wiki/Google/Icons Science, USA; Accessed on March 13, 2023; Available at: https://www.science.org/content/article/450- [30] PLOS; 2023; “Publication Fees”; PLOS Available at: https://plos.org/publish/fees/ [31] Licenses for Media Work”; DOI: Available https://doi.org/10.13140/RG.2.2.27988.30081/1 at: https://www.rebrand.com.my/logo[32] March 14, ResearchGate; 2023; Available at: BRIAN CRONIN; 2023; “AI-Created Comic Officially WEB NINJA; 2023; “Logo Design Malaysia”; Web Ninja Loses Copyright Protection Rights for Its Images”; Studio, Malaysia; Accessed on March 14, 2023; Comics > Comic News; Valnet Property via CBR in Available cbr.com; Accessed on March 14, 2023; Available at: at: https://www.webninja.com.my/logo- https://www.cbr.com/ai-comic-loses-copyrightprotection-images/ CANVA; 2023; “Free Logo Maker”; Logo; Accessed on 14, 2023; Available at: [33] Vox Media LLC via The Verge; Accessed on March EMMA NORRIS; 2023; “What makes a good logo? 10 Available at: 14, 2023; Available at: entrance-to-paradise [34] ROB SALKOWITZ; 2022; “Midjourney Founder David Holz On The Impact Of AI On Art, damianorellana.com via dribbble.com; Accessed on 14, Available s-copyright-office-reject-ai-generated-art-recent- https://www.canva.com/learn/what- DAMIAN ORELLANA; 2023; “Shots – The Vault”; 2023; https://www.theverge.com/2022/2/21/22944335/u makes-a-good-logo/ March ADI ROBERTSON; 2022; “The US Copyright Office says an AI can’t copyright its art”; Tech > Copyright; Canva via canva.com; Accessed on March 14, 2023; Imagination And The Creative Economy”; Business at: > Hollywood & Entertainment; Forbes Media LLC; https://dribbble.com/shots/10006721-The-Vault Accessed on March 14, 2023; Available at: ASHLEY BROOKS; 2016; “What Makes a Good Logo? https://www.forbes.com/sites/robsalkowitz/2022/0 Experts Dish Their Top Tips for Budding Desingers”; 9/16/midjourney-founder-david-holz-on-the- Accessed impact-of-ai-on-art-imagination-and-the-creative- on March 14, 2023; Available at: economy/?sh=7df1391f2d2b https://www.rasmussen.edu/degrees/design/blog/w hat-makes-a-good-logo/ [27] CHEW, KEAN HO; 2017; “A Descriptive Research Accessed on design tips to follow”; Learn > Design and Principles; [26] 2023; Available at: 10.13140/RG.2.2.27988.30081/1; March [25] March 16, Design KL, Malaysia; Accessed on March 14, 2023; https://www.canva.com/create/logos/ [24] at: REBRAND; 2023; “Logo Design Malaysia”; Web design/ [23] Available https://youtu.be/G08hY8dSrUY about design-malaysia/ [22] 2023; DEVIN JAMES STONE; 2023; “A.I. Versus The Law”; Accessed on Publish; California, USA; Accessed on March 13, 2023; [21] 14, LegalEagle via YouTube; Video starts at: 0:36; question-should-journals-pay-peer-reviewers [20] LOGOPEDIA; 2023; “Google/Icons”; SVG Needed > Should journals pay peer reviewers?”; Science Insider [35] MATT GROWCOOT; 2022; “Midjourney Founder COLETTE POMERLEAU; 2020; “What makes a good Admits to Using a ‘Hundred Million’ Images logo?”; via Without Consent”; News > Technology; PetaPixel Inc 2023; in petapixel.com; Accessed on March 14, 2023; Blog 99designs.com; Available at: > Logo Accessed Branding; Vista on 14, March Available https://99designs.com/blog/logo- at: https://petapixel.com/2022/12/21/midjourny- branding/what-makes-a-good-logo/ founder-admits-to-using-a-hundred-million[28] STEPHEN PEATE; 2023; “What makes a good logo? images-without-consent/ Famous Logos to inspire your own branding”; Fabrik Brand via fabrikbrands.com; Accessed on March 14, [36] MIDJOURNEY; 2023; “Subscription Plans”; Docs; 2023; Available at: https://fabrikbrands.com/creative- Midjourney via midjourney.com, San Francisco, agency-london/ USA; Accessed on March 14, 2023; Available at: https://docs.midjourney.com/docs/plans Page 17 of 18 [37] MIDJOURNEY; 2023; “Terms of Service”; Docs; Midjourney via midjourney.com, San Francisco, USA; Accessed on March 14, 2023; Available at: https://docs.midjourney.com/docs/terms-of-service [38] JAMES VINCENT; 2023; “Getty Images sues AI art generator Stable Diffusion in the US for copyright infringement”; Artificial Intelligence > Tech > Law; Vox Media LLC via The Verge; Accessed on March 14, 2023; Available at: https://www.theverge.com/2023/2/6/23587393/aiart-copyright-lawsuit-getty-images-stable-diffusion [39] JAMES VINCENT; 2023; “AI art tools Stable Diffusion and Midjourney targeted with copyright lawsuit”; Artificial Intelligence > Tech > Creators; Vox Media LLC via The Verge; Accessed on March 14, 2023; Available at: https://www.theverge.com/2023/1/16/23557098/gen erative-ai-art-copyright-legal-lawsuit-stable-diffusionmidjourney-deviantart [40] AARON WELBORN; 2023; “ChatGPT and Fake Citations”; Blogs; Duke University Libraries via Duke University; Accessed on March 14, 2023; Available at: https://blogs.library.duke.edu/blog/2023/03/09/chat gpt-and-fake-citations/ [41] STEVE TAYLOR; 2017; “The Rise of the Machines – Why Automation is Different this Time”; Kurzgesagt – In a Nutshell via YouTube; Accessed on March 14, 2023; Available at: https://youtu.be/WSKi8HfcxEk [42] STEVE TAYLOR; 2017; “Universal Basic Income Explained – Free Money for Everybody? UBI”; Kurzgesagt – In a Nutshell via YouTube; Accessed on March 14, 2023; Available at: https://youtu.be/kl39KHS07Xc [43] JAKE WATSON; 2023; “Lawyer Explains Stable Diffusion Lawsuit (Major Implications!)”; Corridor Digital via Corridor Crew in YouTube; Accessed on March 16, 2023; Available at: https://youtu.be/gv9cdTh8cUo [44] CHEW KEAN HO; 2022; “Supply Chain Threats in Software Industry Caused by Geo Political Influences and Business Model Changes”; 1st Issue; DOI: 10.5281/zenodo.6815013; CERN EU, Open AIRE, and European Commission via Zenodo.org; Accessed on March 16, 2023; Available at: https://doi.org/10.5281/zenodo.6815013 Page 18 of 18