OpenAI’s AGI Readiness Czar Resigns, Nvidia Overtakes Apple as World’s Most Valuable Company and Looking At Putting Bugs INTO Neural Networks
Welcome to Hashtag Trending. I’m your host, Jim Love. Let’s get into it.
OpenAI’s AGI Readiness Czar Resigns, Citing Concerns About AI Preparedness
Miles Brundage, OpenAI’s newly-resigned AGI readiness czar, has announced his departure from the organization, expressing deep concerns over the readiness of OpenAI and the world to handle advanced AI technologies. In a personal post on his Substack, Brundage called leaving his “dream job” after six years a difficult decision, driven by a sense of responsibility towards the implications of human-level artificial intelligence.
“I decided,” Brundage wrote, “that I want to impact and influence AI’s development from outside the industry rather than inside.” He emphasized that neither OpenAI nor any other leading lab is prepared for the technology they are developing, and the world is not ready either.
Brundage’s resignation comes amidst a wave of high-profile departures from OpenAI, reflecting broader concerns about the company’s strategic direction and focus. Notable recent exits include Mira Murati, CTO, and Bob McGrew, Chief Research Officer, both of whom left in September 2024 after being instrumental in the company’s operations. Their resignations were followed by Barret Zoph, VP of Research, and Ilya Sutskever, Co-founder & Chief Scientist, who departed in May 2024. Jan Leike, Co-leader of the Superalignment Group, also left shortly after Sutskever, citing concerns about OpenAI’s emphasis on product development over safety. Greg Brockman, OpenAI’s President, is currently on sabbatical, while John Schulman, another Co-founder, recently joined competitor Anthropic.
Brundage was instrumental in creating OpenAI’s five-step scale of AI/AGI levels, which was leaked earlier this year. On this scale, which ends with AI capable of “doing the work of an organization,” OpenAI believes that the world is currently on the verge of level two—AI with human-level reasoning abilities. Despite these advancements, Brundage warned that there are still multiple levels to go, and neither the industry nor society is sufficiently prepared for what comes next.
While Brundage believes AGI has the potential to benefit humanity, he stressed that such outcomes are not guaranteed and will depend on the deliberate actions of those creating and regulating the technology. His resignation comes just months after OpenAI dissolved its AI safety team, and with his exit, OpenAI has reassigned members of its AGI readiness team to other groups, raising further questions about the company’s approach to AGI stewardship
Nvidia Overtakes Apple as World’s Most Valuable Company
Nvidia has overtaken Apple to become the world’s most valuable company, following a record-breaking rally driven by surging demand for its AI chips. On Friday, Nvidia’s market value briefly reached $3.53 trillion, slightly surpassing Apple’s $3.52 trillion, according to LSEG data. Nvidia ended the day valued at $3.47 trillion, while Apple closed at $3.52 trillion.
Nvidia, Apple, and Microsoft have been major contributors to Wall Street’s record highs this year, with their market capitalizations fluctuating closely together. Nvidia, known for its dominance in AI computing chips, has been the biggest beneficiary of the ongoing AI boom, which is seeing increased investments in AI capabilities by companies like Microsoft, Alphabet, and Meta Platforms.
Nvidia’s stock has surged nearly 190% this year, fueled by the growing adoption of generative AI technologies. Its recent rally was further boosted by OpenAI’s $6.6 billion funding round and a strong earnings report from chipmaker TSMC. Nvidia’s chips are in high demand as companies expand their AI initiatives, which has positioned the firm at the forefront of the sector’s growth.
In contrast, Apple is facing challenges with slowing smartphone demand, particularly in China, where iPhone sales slipped by 0.3% in the third quarter. With Nvidia leading the AI race and Apple navigating a competitive smartphone market, the two tech giants continue to drive investor sentiment in the broader market, accounting for a significant portion of the S&P 500 index’s weight.
Space Rover Tests ‘Natural Intelligence’ Based on Insect Brains
For sometime the general wisdom has been that we need larger and larger models for our AI systems to make them truly useful. The problem with that approach is that the costs of training these large and large models is growing exponentially. The compute costs alone just for training GPT4 were estimated at between 60 to 80 million dollars. That’s an increase from the under 5 million dollars reported for GPT 3. And these numbers continue to climb. Google’s next Gemini Ultra is estimated to be closer to 200 million dollars. And these costs don’t include the costs to run these models in computing time and energy usage.
We did a story last week on a firm called LiquidAI as one company that was trying to take a new approach and get more output from a new design that is much smaller and less costly. While the jury is still out on LiquidAI – our first tests had a lot of hallucinations – it’s still very early in its development. But one of the promises of LiquidAI was that it proposed a model so small it could run on a Raspberry PI – the equivalent of a 10 year old laptop.
Will they work through their issues? Probably. But what made them interesting is that they thought differently. Remember that Apple commercial from the 1980’s – “Think Different?”
We here’s some really different thinking.
New intelligent control system models itself on insect brains.
A space rover in a simulated Mars environment is testing a control system inspired by insect brains. The system, developed by UK-based startup Opteran, draws on research from the University of Sheffield, which studied the neurological efficiency of insects such as honeybees.
Opteran’s “Natural Intelligence” aims to replicate the efficiency found in insect brains. Despite having only about 1 million neurons, a bee’s brain can perform complex tasks like navigation, obstacle avoidance, and communication with remarkable energy efficiency. In comparison, large language models (LLMs) like GPT-4 are trained on vast datasets and utilize the computational equivalent of hundreds of billions of parameters, far exceeding the mere 1 million neurons in a bee’s brain. Yet, the efficiency of insect brains offers valuable lessons for creating simpler, energy-efficient control systems. The researchers reverse-engineered these brain functions into software that allows autonomous machines to navigate without extensive data processing or power consumption.
The resulting software, called Opteran Mind, provides space rovers with zero-latency visual depth perception and drastically reduces the need for bulky tools and power-hungry components. It is currently being tested at Airbus Defence and Space’s Mars Yard, with support from the European Space Agency (ESA) and the UK Space Agency. The system’s ability to make near-instant decisions could revolutionize rover operations, allowing them to drive further and faster with minimal human intervention.
Opteran’s ambitions extend beyond space exploration. CEO David Rajan said, “We aim to integrate an Opteran Mind into every machine, underground in mines, on the ground, in the air, and off-world, allowing them to operate as efficiently and as freely as natural creatures.”
It’s not Artificial General Intelligence. But if they get this right, it’s an incredibly efficient computer system that, when compared to other approaches, requires next to nothing in terms of compute cycles or power.
All of that, from intentionally introducing bugs into software discussions. (Sorry, I could not resist).
And that’s our show for today.
Reach me at editorial@technewsday.ca
I’m your host Jim Love, have a Terrific Tuesday