ChatGPT: Beyond The Curious Beast

ChatGPT: Beyond The Curious Beast

ChatGPT, that curious beast, fell into our hands at the end of 2022. Let's take a step back, I said [1].

It's now six months later. And it's become clear, like the observation made by TED conference host Chris Anderson [2], that we need to completely rethink the way we work with a digital interface.

It's an abrupt, sudden change that requires us to adapt to radically new conditions.

? In my lifetime, I’ve seen two demonstrations of technology that struck me as revolutionary. The first time was in 1980, when I was introduced to a graphical user interface ... The second big surprise came just last year [with (Chat)GPT]. ? (Bill Gates) [3]

  1. What we're talking about
  2. What's new and what isn't
  3. What ChatGPT is and what it isn't
  4. Quantum leap
  5. Generative artificial intelligence
  6. The future of artificial intelligence
  7. Speed of change and adaptability

1. What we're talking about

GPT, or Generative Pre-trained Transformer, is an architecture ("Transformer") designed for natural language processing (NLP) -LLM, or Large Language Model, is a natural language processing model-, which has been pre-trained on large sets of textual input data.

This pre-trained architecture - also known as a pre-trained model - is capable of generating new textual data (text, but also computer code or any other language) as output on the basis of a prompt.

You give it a word and it generates the next word, then the word after that, all based on probabilities.

ChatGPT is a practical application of GPT, which is optimised for generating text responses in the context of a conversation.

Architectures similar to Transformer for processing other formats (image, video, audio, etc.) exist or are the subject of active research, for example with video.

2. What's new and what isn't

The Transformer architecture (T of GPT) came out of the Google Brain lab in 2017 [4] [5]. The delivery of NVIDIA's first AI supercomputer to OpenAI [6] and the release of GPT 1 [7] [8] were in 2018. GPT 2, which wowed Geoffrey Hinton [9], one of the leading authorities in the field of artificial intelligence, was released in 2019 [10] [11]. GPT 3 [12] [13] was released in 2020. GPT 4 was ready in August 2022 [14]. And it was on 30 November 2022 [15] that GPT 3.5, an intermediate version, was made available to the general public in the form of a conversational agent (ChatGPT); the update with GPT 4 dates back to March of this year [16]. It is the latter that prompted the open letter [17] calling on ? all AI labs to immediately pause for at least 6 months the training of AI systems more powerful than GPT-4 ?. It should be pointed out that this interruption was addressed to the major players in the sector and not to research as a whole. What's more, some of those with the most advanced models, driven by the wishful thinking of immortality [18] [19], are not necessarily encouraged to support such a moratorium, as the inexorable countdown clock of age continues to tick.

Although he did not sign this letter, Sam Altman, the director of OpenAI, the company that developed ChatGPT, has nevertheless signed an appeal, alongside some 350 experts in artificial intelligence, including Geoffrey Hinton, who described the idea of a moratorium as "silly" [20], to mitigate the "risk of extinction from AI" [21].

This brief history clarifies an important point: it is not the techniques of artificial intelligence, which is a scientific discipline whose aim is to simulate our cognitive functions (perception, reasoning, memory, language), and which were implemented to make ChatGPT possible, that are truly innovative. What is new is the widespread access to the range of possibilities opened up by ChatGPT.

3. What ChatGPT is and what it isn't

ChatGPT is a versatile language model, capable of handling a variety of tasks, both simple and complex. It can perform basic tasks such as sorting a list or finding an email in a pile of text, as well as more advanced tasks such as debugging computer code.

While ChatGPT is an inexhaustible source of information and a great potential breeding ground for ideas, it should be noted that the conversational agent was designed primarily as a tool to aid reasoning [22], making it easier to articulate thoughts.

In factual terms, ChatGPT remains the equivalent of a highly educated pathological liar, capable of communicating both verified facts and lies with equal authority. Its propensity for producing implausible information also makes it an ideal platform for creative expression: ask it to create a far-fetched story about the arrest of an American president wearing Mickey Mouse slippers, and it will. Poem, song, verse in regional dialect; let your imagination run wild. He might even invent a religion [23].

If the hallucinations are supposed to be reduced or even corrected in the short term [3], prefer "ChatGPT helped me to" to "ChatGPT told me", because that is where the strength of this tool lies: iterative interaction. On this point, if you've got your head in the sand, understand that it's not ChatGPT that will threaten your job by automating your tasks [24], but rather those who will use it (with better results than you) to perform those tasks if you don't use it.

What the tool isn't? ChatGPT has no consciousness or will of its own, nor does it have the ability to understand a particular context. While it can't do any of these things, it can or could simulate them. "Simulation" is the key word; let's beware of anthropomorphism. With two million claimed users, including 250,000 who pay for more privacy [25], how many have fallen 'in love' with Replika?

In a nutshell: input data; probability; output data. ChatGPT is nothing more. And that is already a lot: developing an idea, providing contradiction, nuancing a statement. But you can also write a bibliographic reference, correct or improve the level of writing in a text, reformulate a sentence or define a word.

Let's not make the mistake of underestimating ChatGPT's capabilities and the developments we can expect on the pretext that such a tool could never be moved by a sunset: aeroplanes fly faster than birds without feathers, just as computers can solve many problems without feeling or awareness [26].

ChatGPT opens up a field of possibilities that everyone is invited to explore.

FlowGPT is a good starting point for this.

4. Quantum leap

In quantum physics, a quantum jump refers to the phenomenon where a particle passes from one quantum state to another without any apparent transition: the particle "jumps" instantaneously from one state to another, without appearing to pass through intermediate states.

When used figuratively, a quantum leap refers to a sudden and important advance in a particular field, for example a significant change in a given context. The idea is to emphasise the scale and speed of the progress made, as opposed to a gradual and continuous evolution.

While GPT is part of a continuum of advances in artificial intelligence, the fact that access has been extended to as many people as possible with the introduction of ChatGPT has led to a sudden and significant change in our relationship with digital interfaces, a quantum leap of unprecedented magnitude in the digital age: we are moving from driving solo to driving with a co-pilot [27].

This has an immediate consequence: with the (Chat)GPT co-pilot, and by extension with any so-called generative artificial intelligence product, we are sharing and even further delegating cognition and creation to the machine. Carbon and silicon are set to work in harmony.

This new reality gives us a tenfold power of action, in terms of capacity and responsibility: a capacity to do better and more, and an increased responsibility in the face of the capacities of such a tool, now accessible to everyone; ChatGPT, like all technology, is a Pharmakon, both remedy and poison.

5. Generative artificial intelligence

ChatGPT is a practical application of generative artificial intelligence, but there are other examples.

There are services dedicated not only to text (ChatGPT, Claude, Chatsonic), but also to images (DALL-E, Midjourney, Stable Diffusion) and voice (ElevenLabs). Services are also emerging for video (Runaway) and music (MusicLM). These few examples represent just a small sample of a list that continues to grow daily.

In the image category alone, the current use of Midjourney [28] makes it one of the most striking artistic expressions in the history of humanity. A new Renaissance? That is the idea shared by Yann LeCun about the potential of artificial intelligence [29]. The prompthero.com website and the page dedicated to #Midjourney on Instagram provide evidence to support this idea.

6. The future of artificial intelligence

Whereas the power of artificial intelligence was previously confined to the realm of laboratories, it is now possible for anyone to use ChatGPT and its derivatives, prompting reactions from some of the leading figures in the field of artificial intelligence, as well as from the economic and political spheres.

Like the Covid-19 pandemic, the future of artificial intelligence in the GPT era is generating alarmist positions among some and reassuring positions among others.

For some, it is no longer a question of a virus as a potential threat to human life, but of technology as a potential threat to all human life.

Geoffrey Hinton believes that we have found the secret of immortality, but that it does not concern us [30]. He suggests that human intelligence may only be a transitional stage in the evolution of intelligence [20].

Max Tegmark of MIT even worries about an AI "Don't look up" [31], in reference to the tragicomic film in which an asteroid is heading towards the Earth to destroy it. Tristan Harris, like Max Tegmark in the aforementioned article, points out that 50% of AI researchers believe there is a 10% or greater probability that AI will lead to the extinction of humanity, citing a survey conducted in the summer of 2022 [32]. This argument was used to alert public opinion in a conference entitled "The A.I. Dilemma", which has been viewed more than two million times on YouTube.

As for Sundar Pichai, current boss of Alphabet and Google, he expressed his concern to CBS News [33] about the fact that Bard, Google's ChatGPT, was able to learn Bengali without having been specifically trained to do so. These are emergent properties.

Although concerns about the future implications of developments in artificial intelligence are legitimate - nukes don't make stronger nukers, whereas AI makes stronger AI [34] - they need to be qualified: it is not half of artificial intelligence researchers who are worried about a threat of extinction, but 162 of the 738 researchers surveyed who agreed to respond, out of a sample of 4,271 people, i.e. 4% of the sample [35]. As for learning Bengali, Margaret Mitchell, a former Google employee, disputes this claim [36], and a recent study by Stanford University [37] proposes a completely different explanation for emergent properties.

For others, such as Jean-Gabriel Ganascia, there is no need to overreact, as there is no scenario in which the machine could automatically switch on and take control [38].

However, Jean-Gabriel Ganascia's short-term vision seems to overlook what NVIDIA CEO Jensen Huang sees as the next step: the integration of AI into the physical world [39]. The investment by OpenAI [40] [41], which closed its robotics division in 2021 to focus on what will become ChatGPT [42], in Norwegian robotics company 1X, which specialises in making "general purpose androids, inspired by human nature", is a step in this direction. The investment is aimed in particular at bolstering efforts to build the next model of bipedal android called NEO, which aims to "explore how artificial intelligence can take form in a human-like body".

Whatever the position taken by the various parties, there seems to be a consensus on one aspect of the debate: it is essential to rigorously address the question of the future of artificial intelligence, and to do so without delay.

7. Speed of change and adaptability

Astro Teller is best known for his eponymous graph. In essence, the idea is as follows: scientific and technological progress is now faster than our ability to adapt to its consequences.

The Center for Humane Technology illustrates this [43] by pointing out that we have not finished adapting our society to the consequences of a technology as simple as a social network, and we must already take into account the issues linked to widespread access to the capabilities of artificial intelligence.

No alt text provided for this image

With ChatGPT and similar generative cohorts, we have given free rein to dizzying power, which is causing concern among many players.

Eric Schmidt, who was head of Google for a decade (2001-2011) and has been active in the technology sector for half a century, admits himself [44] that he has never experienced such an acceleration of innovation since the public roll-out of ChatGPT and that he is struggling to keep up.

A non-exhaustive list of recent announcements gives an idea of this lightning acceleration: AI Sandbox (Meta), Bard (Google), Bedrock (Amazon), Bing Chat (Microsoft), MyAI (Snapchat), Slack GPT (Salesforce).

Faced with these tools, which have an extremely high proliferation coefficient [45], we are faced with a change whose speed is worrying even the head of OpenAI [46].

In essence, speed is OpenAI's argument to justify the public deployment of ChatGPT, despite certain misgivings: make the power of GPT accessible now, when the stakes, which are already high and complex to manage, remain relatively limited.

In short, making GPT accessible now is in some ways a way of cushioning the social and societal shock it will cause.

Whatever the approach adopted and however dizzying the speed of change, the point of convergence remains the same: our humanity and its future.

Bibliography

[1] Monde Numérique (podcast). (19 December 2022). ChatGPT : une révolution en 5 actes (Denys Malengreau, éditorialiste numérique) [Online video]. YouTube. Retrieved from https://youtu.be/xaBIwQ6zgsI?t=99

[2] TED. (20 April 2023). "The Inside Story of ChatGPT's Astonishing Potential" [Video online]. YouTube. Retrieved from https://youtu.be/C_78DM8fG6E?t=1006

[3] Gates, B. (21 March 2023). The Age of AI has begun [Article online]. Notes. Retrieved from https://www.gatesnotes.com/The-Age-of-AI-Has-Begun

[4] Uszkoreit, J. (31 August 2017). Transformer: A Novel Neural Network Architecture for Language Understanding [Article online]. Google AI Blog. Retrieved from https://ai.googleblog.com/2017/08/transformer-novel-neural-network.html

[5] Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (June 12, 2017). Attention Is All You Need [Article online]. arXiv preprint arXiv:1706.03762. Retrieved from https://arxiv/abs/1706.03762

[6] CNBC (19 March 2023). CEO Jensen Huang On How His Big Bet On A.I. Is Finally Paying Off - Full Interview [Video online]. YouTube. Retrieved from https://youtu.be/_-wjA4XlBl4?t=844

[7] OpenAI. (11 June 2018). Improving language understanding with unsupervised learning [Article online]. Retrieved May 10, 2023, from https://openai.com/research/language-unsupervised

[8] Radford, A., Narasimhan, K., Salimans, T., & Sutskever, I. (2018). Improving Language Understanding by Generative Pre-Training [Article online]. OpenAI. Retrieved from https://cdn.openai.com/research-covers/language-unsupervised/language_understanding_paper.pdf

[9] CBS Mornings (25 March 2023). Full interview: "Godfather of artificial intelligence" talks impact and potential of AI [Video online]. YouTube. Retrieved from https://youtu.be/qpoRO378qRY?t=42

[10] OpenAI. (14 February 2019). Better language models and their implications [Article online]. Retrieved from https://openai.com/research/better-language-models

[11] Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., & Sutskever, I. (2019). Language Models are Unsupervised Multitask Learners [Article online]. OpenAI. Retrieved from https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf

[12] Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., ... & Amodei, D. (28 May 2020). Language models are few-shot learners [Article en ligne]. OpenAI. Retrieved from https://openai.com/research/language-models-are-few-shot-learners

[13] Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., ... & Amodei, D. (28 May 2020). Language models are few-shot learners [Article en ligne]. OpenAI. Retrieved from https://openai.com/research/language-models-are-few-shot-learners

[14] ABC News. (March 18, 2023). OpenAI CEO, CTO on risks and how AI will reshape society [Online video]. YouTube. Retrieved from https://youtu.be/540vzMlf-54?t=783

[15] OpenAI. (30 November 2022). Introducing ChatGPT. Retrieved from https://openai.com/blog/chatgpt

[16] OpenAI. (14 March 2023). GPT-4. Retrieved from https://openai.com/research/gpt-4

[17] Future of Life Institute. (22 March 2023). Pause Giant AI Experiments: An Open Letter [Article online]. Retrieved from https://futureoflife.org/open-letter/pause-giant-ai-experiments

[18] Bloom, E. (31 March 2017). Google's co-founders and other Silicon Valley billionaires are trying to live forever. CNBC. Retrieved from https://www.cnbc.com/2017/03/31/google-co-founders-and-silicon-valley-billionaires-try-to-live-forever.html

[19] Regalado, A. (8 March 2023). Sam Altman invested $180 million into a company trying to delay death. MIT Technology Review. Retrieved from https://www.technologyreview.com/2023/03/08/1069523/sam-altman-investment-180-million-retro-biosciences-longevity-death

[20] Mearian, L. (4 May 2023). Q&A: Google's Geoffrey Hinton - humanity just a 'passing phase' in the evolution of intelligence. Computerworld. Retrieved from https://www.computerworld.com/article/3695568/qa-googles-geoffrey-hinton-humanity-just-a-passing-phase-in-the-evolution-of-intelligence.html

[21] Amodei, D., Altman, S., Bengio, Y., Gates, B., Hassabis, D., Hinton, G., Song, D., et al. (30 May 2023). "Statement on AI Risk." Retrieved from https://www.safe.ai/statement-on-ai-risk#open-letter

[22] ABC News. (March 18, 2023). OpenAI CEO, CTO on risks and how AI will reshape society [Video online]. YouTube. Retrieved from https://youtu.be/540vzMlf-54?t=627

[23] Howey, H.; Dupont-Besnard, M. (10 April 2023). ChatGPT invente une nouvelle religion avec 10 commandements [Article online]. Numerama. Retrieved from https://www.numerama.com/tech/1333586-chatgpt-invente-une-nouvelle-religion-avec-10-commandements.html

[24] Schell, H. (30 April 2023). Which Jobs Will Be Most Impacted by ChatGPT? Visual Capitalist. Retrieved from https://www.visualcapitalist.com/cp/which-jobs-artificial-intelligence-gpt-impact

[25] Tong, A. (21 March 2023). What happens when your AI chatbot stops loving you back? [Online article]. Reuters. Retrieved from https://www.reuters.com/technology/what-happens-when-your-ai-chatbot-stops-loving-you-back-2023-03-18

[26] Alexandre, L. (2023). La guerre des intelligences à l'heure de ChatGPT. JC Lattès, p. 220.

[27] Microsoft Developer (23 May 2023). Announcing Windows Copilot [Video online]. YouTube. Retrieved from https://youtu.be/FCfwc-NNo30

[28] SimilarWeb. Retrieved from https://www.similarweb.com/website/midjourney.com/#traffic

[29] DeepLearningAI. (7 April 2023). Yann LeCun and Andrew Ng: Why the 6-month AI Pause is a Bad Idea [Video online]. YouTube. Retrieved from https://www.youtube.com/live/BY9KV8uCtj4?feature=share&t=1007

[30] Hern, A. (5 May 2023). We've discovered the secret of immortality. The bad news is it's not for us': why the godfather of AI fears for humanity. The Guardian. Retrieved from https://www.theguardian.com/technology/2023/may/05/geoffrey-hinton-godfather-of-ai-fears-for-humanity

[31] Tegmark, M. (25 April 2023). The 'Don't Look Up' Thinking That Could Doom Us With AI. TIME. Retrieved from https://time.com/6273743/thinking-that-could-doom-us-with-ai

[32] AI Impacts. (August 3, 2022). 2022 Expert Survey on Progress in AI. Retrieved from https://aiimpacts.org/2022-expert-survey-on-progress-in-ai

[33] CBS News. (16 April 2023). The AI revolution: Google's artificial intelligence developers on what's next in the field. 60 Minutes Overtime. Retrieved from https://www.cbsnews.com/news/ai-advancements-google-artificial-intelligence-future-60-minutes-2023-04-16

[34] Center for Humane Technology (5 April 2023). The A.I. Dilemma - March 9, 2023 [Video online]. YouTube. Retrieved from https://youtu.be/xoVJKj8lcNQ?t=2277

[35] Sundin, M. (9 May 2023). No, 50% of AI researchers don't believe there is a 10% risk that AI will kill us. Warp News. Retrieved from https://www.warpnews.org/artificial-intelligence/no-50-of-ai-researchers-dont-believe-there-is-a-10-risk-that-ai-will-kill-us

[36] Mitchell, M. [@mmitchell_ai]. (17 April 2023). Tweet [Status]. Twitter. Retrieved from https://twitter.com/mmitchell_ai/status/1648029417497853953

[37] Xiang, C. (2 May 2023). Scary 'Emergent' AI Abilities Are Just a 'Mirage' Produced by Researchers, Stanford Study Says. Vice. Retrieved from https://www.vice.com/en/article/wxjdg5/scary-emergent-ai-abilities-are-just-a-mirage-produced-by-researchers-stanford-study-says

[38] éLUCID. (6 May 2023). LA FACE CACHéE DE L’IA : QUI CONTR?LE LES ROBOTS ET DANS QUEL BUT ? - Jean-Gabriel Ganascia [Video online]. YouTube. Retrieved from https://youtu.be/sHDKaS3GQN4?t=2718

[39] CNBC (19 March 2023). Nvidia CEO Jensen Huang On How His Big Bet On A.I. Is Finally Paying Off - Full Interview [Video online]. YouTube. Retrieved from https://youtu.be/_-wjA4XlBl4?t=1904

[40] 1X (23 March 2023). 1X Raises $23.5M in Series A2 Funding led by OpenAI. Retrieved from https://1xtech.medium.com/1x-raises-23-5m-in-series-a2-funding-led-by-openai-6040af4f3f4f

[41] Blain, L. (10 April 2023). OpenAI and Figure join the race to humanoid robot workers. New Atlas. Retrieved from https://newatlas.com/robotics/openai-figure-ai-robotics

[42] Wiggers, K. (16 July 2021). OpenAI disbands its robotics research team. VentureBeat. Retrieved from https://venturebeat.com/business/openai-disbands-its-robotics-research-team

[43] Center for Humane Technology. 5 April 2023. The A.I. Dilemma [Video online]. YouTube. Retrieved from https://youtu.be/xoVJKj8lcNQ?t=2567

[44] ABC News. (April 2, 2023). This Week' Transcript 4-2-23: Joe Tacopina, Former Gov. Asa Hutchinson and Eric Schmidt [Online transcript]. Retrieved from https://abcnews.go.com/Politics/week-transcript-4-2-23-joe-tacopina-former/story?id=98208

[45] France Culture. (4 April 2023). ChatGPT, Midjourney : pourquoi les experts de l'IA appellent à ralentir [Video online]. YouTube. Retrieved from https://youtu.be/XiL1dgsYN5g?t=128

[46] ABC News. (18 March 2023). OpenAI CEO, CTO on risks and how AI will reshape society [Video online]. YouTube. Retrieved from https://www.youtube.com/watch?v=540vzMlf-54&t=715s

Maya Nix

Building brands & driving business growth with Storytelling | Creative Copywriting & Content Consulting Services

1 年

Well done, Denys! A thorough, and responsible, layout of Chat GPT. Check out this article about the potential problem for training generative AI in future, as more and more AI-generated text and synthetic data is published online: https://cosmosmagazine.com/technology/ai/training-ai-models-on-machine-generated-data-leads-to-model-collapse/

回复

要查看或添加评论,请登录

社区洞察

其他会员也浏览了