.fluid-width-video-wrapper{left:0% !important;width:100%!important;position:relative !important;}
In the tech and software industry, product and service iterations happen incredibly quickly, it’s inherent in the process of iterative design and fast software cycles. We’ve become accustomed to next-day changes and daily patch notes. However, nothing could have prepared us for the number of changes that would take place at OpenAI as a company this past week (20th – 24th November 2023). To try and understand what happened, it’s important to understand the history of OpenAI and Sam Altman’s role in the company.
How far away is AGI? It may be closer than we think
Let’s dive into the where, the what, and the when, and chart the history of OpenAI and its turbulent road ahead. This is a road that could not only transform an entire industry but create one that has the potential to change humanity forever.
OpenAI’s history – A note about discovery, innovation, and implication
Before we really get into the future of OpenAI and their recent ‘edit > undo’ in regards to Sam Altman, let’s take a moment to focus on the work that has been done on ChatGPT, AI (Artificial Intelligence), AGI (Artificial General Intelligence) and the history of OpenAI so far.
It may sound dramatic and grandiose, but OpenAI’s products stand to completely transform nearly every industry, impacting every person on the planet.
Even those who aren’t glued to their screens asking ChatGPT questions like ‘what would you do if you were sentient?’, everyone on the planet may be impacted by the eventual evolution of AI into AGI and potentially even ASI. In the following link, we covered in detail what is meant by true AI and the differences between AI, AGI and ASI:
We’ll touch upon it here but the article is well worth a read to understand these terms and how they relate to recent events at OpenAI.
Many of humanity’s most significant breakthroughs came as a result of curiosity-driven science, and the whole field of AI and AGI may have similar implications.
We didn’t set out to discover electromagnetism
We didn’t set out to discover radio technology
We didn’t intend to discover penicillin
These breakthroughs were discovered whilst on the road to researching something entirely unrelated, or in some cases, completely by accident.
History tends to repeat itself and leaves us wondering what profound discoveries will come next in a similar fashion. Could the history of OpenAI and its recent journey facilitate such discoveries? From improved medical research and quicker breakthroughs to helping us answer the most fundamental questions humans have asked since the beginning of time, this may be what the not-too-distant future of OpenAI and AI, in general, has in store for us.
But isn’t what we think of as AI just an LLM (Large Language Model)?
For those who are thinking this after reading the previous paragraph, you are completely right.
As it stands at the moment, ChatGPT is ‘just’ an LLM and uses a prediction model. You can learn more about LLMs and how ChatGPT works at the following link:
How Does ChatGPT Work in Story Form: The Adventures of ChatGPT “The AI Storyteller”
Although neither ChatGPT nor LLMs constitute AGI or even true intelligence, we should be open to the idea of AGI happening as a possible result of combining many learning and language models in the future.
For those who break down the technology and systems that make up a LLM like ChatGPT, they’ll most likely conclude that they just consist of code running on a server. You could just as easily trivialise the human brain by stating that it’s just a bunch of cells and electrical impulses. My point is, that we need to think from the perspective of these systems being greater than the sum of their parts, so to speak.
LLMs and the parts that make up AI, we may be closer to AGI than thought
Although we aren’t at a point yet where the code/parts are clever enough to achieve AGI (at least not publically), we may not be far away from that becoming the case.
The history of Open AI – the story so far…
OpenAI has seemingly appeared out of nowhere, going from being a relatively unheard-of company a number of years ago, to being a constant hot topic in the media.
The incredible events of 2023 are the result of the entire history of OpenAI and the decisions made, but before looking at that history, let’s pause and think about 2023.
Now – OpenAI present (2022 to 2023)
For those interested in technology and even those who aren’t, it’s hard to argue that this year has seen a huge change for AI but also a technology inflection point that will go down in history as a year to rival the internet in 1983:
But how did this happen? Here’s a brief timeline of events:
- On 30th November 2022, ChatGPT was released, making “AI” available to the public (for free)
- 5 days later, ChatGPT reached 1 million users, overtaking every other app in history
- By February 2023, only three months later, ChatGPT sets the record as having the fastest-growing growing user-base ever (although later overtaken later by Meta’s Threads release on 5th June 2023)
- By April 2023, OpenAI’s founder Sam Altman was named one of the top 100 most influential people of 2023 by Time Magazine
- During 2023 alone, we’ve seen major, almost unrepentant advancements in what ChatGPT can do, e.g.
- Voice Availability (Nov 21, 2023)
- Custom GPTs (Nov 6, 2023)
- Browsing Out of Beta, “Browse with Bing” in GPT-4 (Oct 17, 2023)
- DALL·E 3 Integration For Image Generation (Oct 16, 2023)
- Browsing for Plus Users (Sep 27, 2023)
- New Voice and Image Capabilities (Sep 25, 2023)
- Multilingual Support – Alpha (Sep 11, 2023)
- Custom Instructions for Free Users (Aug 9, 2023)
- Higher Message Limits for GPT-4 (July 19, 2023)
- Code Interpreter Beta (July 6, 2023)
- ChatGPT App for iOS and Android (May to July)
- Web Browsing and Plugins Beta (May 12, 2023)
- ChatGPT Plugins Introduction (March 23, 2023)
- GPT-4 Introduction for Plus Users (March 14, 2023)
- Introduction of ChatGPT Plus (Feb 9, 2023)
- Early 2023 Updates (Jan 9, 2023)
- Then on 17th November 2023 (days after the rollout of GPTs), Sam Altman was fired and re-hired on November 21st 2023. More on this later!
We’ve seen crazy things happen in the world of Silicon Valley, but nothing even comes close to this.
So let’s sit back and understand how the history of OpenAI led to the events of 2023.
OpenAI was founded (2015)
It may surprise you to learn that Elon Musk was one of the founders of OpenAI along with Sam Altman, only to leave OpenAI and later invent his own AI bot to rival ChatGPT called Grok.
OpenAI Gym (2016)
OpenAI released its first major project, the OpenAI Gym for reinforcement learning research.
This is a toolkit that is designed to serve as a platform for researchers and developers to experiment and benchmark their reinforcement learning algorithms.
This was introduced as an open-source project, showing OpenAI’s commitment to openness and collaboration in the field of AI.
OpenAI Five (2017 & 2018) and my first encounter of OpenAI at the Dota 2 international tournament
My first personal experience of OpenAI was back in 2017 while watching an eSports tournament. This was the Dota 2 International – an event where the best teams go head-to-head for a large multi-million dollar prize pool.
It is perhaps useful to preface this by saying that Dota 2 is a video game that has been described as a combination of American football and chess. It’s a complex online game with many different situations, scenarios, and synergies that can arise in a battle of 5v5.
At the Dota 2 international event, an AI was trained to compete 1v1 with professional Dota 2 players which resulted in some surprising and shocking demonstrations of the AI’s ability to learn and adapt.
AI bots trained to learn the game
It was later revealed that while OpenAI bots were ‘learning’ the game, the simulations performed daily would allow for 300 years worth of Dota game experience to be simulated within 24 hours.
There have since been numerous examples of OpenAI outplaying their human opponents.
Two professional Dota 2 players react to OpenAI’s in-game ability
After OpenAI’s first appearance at the Dota 2 international, the team continued to work on the model which resulted in a team of 5 AI that could hold their own against professional teams. In some cases even winning. This AI team became OpenAI Five.
There is more information on the official OpenAI Five website and it’s a fascinating read, even for non-gamers.
GPT-2 (2019)
GPT-2 was released in February and was trained on a dataset of over 7,000 books and 8 million web pages.
This update to GPT allowed for full paragraphs of text with comprehensive replies to the user.
Despite being released in February 2019, it wasn’t available to general users until later in the year due to concerns over misuse. The delay was likely due to the team building a way to deal with the misuse potential of the upgrade which is understandable.
GPT-3 (2020)
This was the moment when things changed for OpenAI. The natural language aspect of the model’s improvements made it very believable and authentic for portraying convincing human-like conversations.
The API was also released allowing third-party developers to make their own intelligent applications. Businesses found the idea of GPT-type chat support agents to improve their troubleshooting processes and general customer support.
Other use cases range from using GPT-3 as a coding assistant to content creation and language translation services.
GPT-4 & introduction of ChatGPT Plus a.k.a. ChatGPT-4 (2023)
As it stands at the time of writing (November 29th 2023), Chat GPT-4 is only available to those who are subscribers to ChatGPT Plus.
ChatGPT Plus was introduced around the same time as the GPT-4 model, which was a big incentive for fans of GPT-3 who wanted to get their hands on the latest version of GPT-4.
The full timeline of events can be found in our opening section above, although it’s important to mention here that ChatGPT is just one of OpenAI’s offerings.
Before ChatGPT Plus, we only had the free version, and at that time, OpenAI were trying to assess what users would pay for a premium version of ChatGPT.
However, users were also able to sign up OpenAI directly to access their playground and APIs for development using credits. You can find out more about the differences between ChatGPT and OpenAI’s playground here, but from recollection, we had access to the GPT-4 model slightly earlier using the playground, allowing users to easily switch between older models and the very latest model.
The firing and reappointment of Sam Altman
The firing of Sam Altman sent shockwaves through the tech industry last week leaving many users and media outlets confused. Conspiracy theories and rumours were running wild the weekend of the firing leaving it almost impossible for followers to know what truly happened.
Initial reports stated that Altman had been employed by Microsoft, with other reports flying around, including Charles Hoskinson from crypto’s Cardano project inviting Sam Altman to do a partnership. To say the situation was confusing would be an understatement.
Not only was it impossible to know what had happened, or where Sam Altman was going, the biggest question was “why?”.
The man named one of the top 100 most influential people by Times Magazine being fired by the board only days after one of the biggest updates to ChatGPT (custom GPTs), it made no sense at all. I personally thought it was related to GPTs somehow given how close the two events happened, but was most likely wrong.
By Tuesday 21st November, only days after the firing, Sam Altman returned to OpenAI as CEO.
Events surrounding Sam Altman’s firing
Let’s take a closer look at the events:
- November 17th – OpenAI board announces the firing of Sam Altman. Later the same day, Greg Brockman (OpenAI co-founder) resigned. Many more employees also announced their resignation shortly after.
- November 18th – Pressure builds from investors unhappy with the recent firing and resignations of key team members of the company. Microsoft had recently promised to invest $13 billion into OpenAI just a short few months ago. Members of the team were assured that Altman and Brockman would be returning. After numerous employees were voicing their concerns at the recent changes.
- November 19th – Sam Altman returned to OpenAI offices. It was reported that Microsoft’s CEO (Satya Nadella) was an integral part of pushing for Altman’s return to OpenAI. More confusion came later as OpenAI announced a new interim CEO (Emmett Shear).
- November 20th – Microsoft announces that Altman, Brockman and other former employees from OpenAI would be joining Microsoft, adding to the confusion amongst investors and the community. This seemed to pour more fuel on the fire as even more OpenAI employees were threatening to quit on the spot. Later that day, the co-founder of OpenAI (Ilya Sutskever) signed a petition along with more employees, calling for the board to step down.
- November 21st – Sam Altman was officially back as CEO of OpenAI, with all but one of the previous board of directors remaining. For more on who was in and out, you can get the lowdown here.
After such a rollercoaster that wouldn’t look out of place on an episode of Silicon Valley, we still don’t know 100% why Altman was fired but reports seem to be firming up something that (if true) is remarkable.
AGI & Q* (Q-Star): The real reason for Altman’s firing
While it initially seemed as though Altman’s firing might be related to the release of GPTs, two days after the sacking of the re-hiring of Sam Altman, news started to circulate that Sam’s firing was related to “AI safety concerns” surrounding a breakthrough internally at OpenAI.
As reported by The Guardian here, along with various other sources including The Independent, Routers, and others, it seems that a new AI model, referred to as Q* (Q-Star), led to the temporary ousting of Sam Altman. OpenAI researchers sent a letter to the board, expressing concern about the potential dangers of this powerful AI discovery.
The model, which showed promise in solving mathematical problems, raised optimism about achieving AGI, with researchers believing that mastering math, a field where there’s usually only one correct answer, would indicate an AGI-level of advancement in the AI’s reasoning capabilities.
Key points surrounding Q* and why Sam Altman was fired
- Development of Q* (Q-Star): OpenAI was reportedly developing an advanced AI model called Q*, which demonstrated the ability to solve unseen basic maths problems, indicating progress towards AGI. Mastery of math by AI suggests advanced reasoning capabilities, akin to human intelligence.
- Safety concerns and staff reaction: The power and potential of Q* alarmed some OpenAI researchers, prompting them to write to the board about the potential threats to humanity. This concern was a key factor in the initial dismissal of CEO Sam Altman.
- Sam Altman’s ouster and reinstatement: Following the internal concerns and the letter to the board, Sam Altman was removed from his position. However, his dismissal led to a strong reaction from the staff, with threats of mass resignations from over 700 employees. This reaction, coupled with support from Microsoft, OpenAI’s major investor, led to his reinstatement.
- The Quest for Artificial General Intelligence (AGI): OpenAI’s work on Q* is part of its broader goal to achieve AGI, which refers to AI systems that can perform a wide variety of tasks at human or superior levels of intelligence.
- Public and Internal Responses: The situation has sparked discussions about the pace of AI development and the ethical implications of rapidly advancing AI technology. Internally, the company acknowledged the project and the letter but did not comment on the specifics.
- Altman’s Vision and Progress: Despite the turmoil, Altman has been a key figure in pushing OpenAI towards significant advancements in AI, including the development of ChatGPT, one of the fastest-growing software applications in history.
Our view: this was already on the cards with Jimmy Apples AGI predictions
What’s most interesting here is that we produced our AGI article originally on 21st June 2023 and predicted that AGI was just around the corner despite many believing we were years if not decades away from achieving AGI. On 2nd October we gave the article an update, theorising that AGI had already been developed internally at OpenAI.
At this point, there were no solid reports of this from outlets like The Guardian or The Independent, but mysterious claims from somebody called Jimmy Apples appeared online, somebody who had a track record of making accurate predictions about OpenAI and ChatGPT.
The claims made by Jimmy Apples implied that OpenAI were delaying the public release of their AGI model. After making these claims, Jimmy Apples deleted his Twitter and Reddit accounts and vanished entirely from social media.
What’s even funnier is that Sam Altman himself reported about AGI being a potential co-worker replacement and claiming on Reddit that:
“AGI has been achieved internally.”
But then claims:
“Obviously this is just memeing, y’all have no chill, when AGI is achieved it will not be announced with a Reddit comment.”
Viewing these events and comments in relation to the most recent Q* announcements, it seems hard to dismiss that they are all related.
What’s next for the future of OpenAI?
We’ve covered the history of OpenAI, recent events around Sam Altman’s firing and rehiring, and how this relates to AGI. So, what does all this mean for the future of OpenAI?
The death of OpenAI
It’s hard to imagine now, but with the very real prospect of the founder and CEO, Sam Altman, Greg Brockman (co-founder), and potentially the entire workforce offering to leave and join Microsoft, this could have quite easily led to the death of OpenAI, or at least as we know it. It’s almost impossible to imagine the future of OpenAI without Sam Altman, and even more impossible to imagine without the 700 or more employees who offered to quit.
The thought that’s left in my mind, is whether what happened was just a precursor to future events, leading to the actual death of OpenAI and reinvention as something else. Time will tell, but for now, OpenAI is still alive and well with Sam at the helm.
GPT-5 AGI
This was discussed in our original AGI article but given its relevance here, I’ve included a list of FAQs at the very end that will help to answer any questions relating to AGI.
A vision of the future of OpenAI
Competition is heating up
Elon Musk’s Grok
Elon Musk’s Grok AI has been seeing a lot of attention as of late and is set to challenge ChatGPT as the go-to LLM solution with its ability to access real-time Twitter (X) information:
In a recent podcast with Lex Fridman, Musk put Grok through its paces, demonstrating the more light-hearted approach to serving enquiries with less emphasis on censorship and restrictions that many have criticised GPT for.
Elon and Lex test out Grok AI on fun mode
At the time of writing, Grok is only available to verified users of X but is expected to roll out to a wider audience in the coming months.
Google’s Bard
Meanwhile, Google Bard has begun integration into YouTube, allowing users to request the LLM to give a rundown or brief description of video content.
Bard still trails behind OpenAI’s ChatGPT, but the gap is expected to close further within the next 6-12 months in terms of the capability of currently available LLMs.
Anthropic’s Claude
Despite Google’s own ventures into AI with Bard and AI-first search, Google is also a significant investor of Anthropic presumably to rival Microsoft’s investment in OpenAI.
AI Means The Death of SEO? Not So Fast – Let’s Look At How AI is Shaping The Future of SEO
I’ve tested Anthropic’s Claude 2 model with its massive 100K token limit. It’s good, but similar to Bard, trails behind ChatGPT, failing to offer the same experience or degree of innovation. That said, its 100K token limit is a major bonus, with its ability to process vast amounts of information (over three times the amount of GPT-4’s 32K model).
As with anything in the world of AI, it’s a competition. Earlier this month, OpenAI announced GPT-4 Turbo with a 128K model, only to be outdone again by Anthropic who announced Claude 2.1 with an incredible 200K token limit.
Conclusion
So much has changed in the AI and LLM space but one thing has been constant – it’s such a new, exciting and transformative set of technologies.
These technologies stand to increase efficiency for businesses across multiple industries and one player stands out head and shoulders above all others right now, OpenAI.
Other than the dangers of advanced AI and AGI, one concern is the eventual stagnation of the technology due to filters and limitations imposed on the system to stop offensive content. It would be such a shame to see these kinds of limitations delay or even prevent progress.
It will be an exciting few months following the upcoming developments from OpenAI and its competitors.
What do you think about the recent events at OpenAI? Have we covered the history and future of OpenAI adequately? Leave us a comment in the comment section below or join us on social media.
0 Comments