In the aftermath of last week’s shocking OpenAI power strugglethere was a final revelation that served as a kind of epilogue to the vast chaos. report An article from Reuters revealing perhaps surprising advances in startups. The breakthrough is said to have happened through a little-known program called “Q-Star” or “Q*.”
According to the report, this Q-related “discovery” was one of the things that may have triggered internal conflict at the influential AI company. Prior to Altman’s firing, several OpenAI staffers allegedly sent a letter to the company’s board about “discoveries of powerful artificial intelligence that they claim could threaten humanity.” Reuters, citing unnamed sources, claimed the letter was “one of a long list of board complaints that led to Mr. Altman’s dismissal.”
Frankly, that story sounded pretty crazy. What was this strange new program? Why is it thought to have caused havoc at OpenAI? Reuters reports that the Q* program allows AI agents to perform “grade-level math” If true, this is an amazing technological advance that could be a huge success in the development of artificial general intelligence (AGI), sources said.another report The Information article largely repeats many of the points made in the Reuters article.
Still, no details about this supposed Q program have been shared by the company, leaving only reports from anonymous sources and rampant speculation online as to what the true nature of the program is.
Some have speculated that the program (given its name) may be related to Q-learning, a type of machine learning. What is Q-learning? How can it be applied to OpenAI’s secret programs?
In general, there are several different ways to teach an AI program to do something.One of them is known as “Supervised learning”which works by feeding AI agents with large tranches. “Labeled” data, which is then used to train the program to perform a function on its own (usually that function is a more detailed data classification). Generally speaking, programs like ChatGPT, OpenAI’s content generation bot, were created. Use some form of supervised learning.
Unsupervised learning, on the other hand, is a type of ML in which an AI algorithm can sift through large portions of unlabeled data to find patterns to classify. This type of artificial intelligence can be deployed for a variety of purposes. Recommendation system Companies like Netflix and Spotify use it to suggest new content to users based on past consumer choices.
Finally, there is reinforcement learning (RL). This is a category of ML that prompts an AI program to achieve a goal within a specific environment. Q-learning is a subcategory of reinforcement learning. In RL, the researcher treats her AI agent like a dog in training. Programs receive “rewards” for performing certain actions to influence specific outcomes, and are penalized for performing other actions. In this way, the program is effectively “trained” to pursue the most optimized outcome in a given situation. In Q-learning, the agent obviously uses trial and error to find the best way to achieve the goals it is programmed to pursue.
What does this all have to do with OpenAI’s supposed “mathematical” breakthroughs? Programs that can (supposedly) perform simple mathematical operations can access that ability via some form of Q-related RL. One might speculate that it may have reached . However, many experts are somewhat skeptical about whether AI programs can actually solve math problems. Even if AI could achieve such a goal, some people seem to think it is impossible. Not necessarily leading to widespread AGI breakthroughs. MIT Technology Reviews, Interviews:
For years, researchers have been trying to obtain AI models to solve mathematical problems. Language models such as ChatGPT and GPT-4 can perform some computations, but they are not very good or reliable. Wenda Lee, her AI lecturer at the University of Edinburgh, says that we currently don’t have the algorithms or even the right architecture to reliably solve math problems using AI. The deep learning and transformers (a type of neural network) that language models use are good at recognizing patterns, but that’s probably not enough, she adds.
In short, we don’t really know much about Q, but if the experts are to be believed, the hype around it might just be hype.
Question of the day: Seriously, what the hell happened to Sam Altman?
It’s worth some consideration that we still don’t know what the hell happened to Sam Altman last week, despite the fact that he’s back at OpenAI.in interview In an interview with The Verge on Wednesday, Altman said little about what sparked a dramatic power struggle at his company last week. Despite continued grilling from reporters at the outlet, Altman held up his hands and said he would not discuss the matter for the time being. “I understand why people want answers right now. But I also think it’s completely unreasonable to expect that,” the resilient CEO said. Rather, the most the Verge could glean from OpenAI executives was that the company was conducting an “independent investigation” into what happened — a process he said He says he doesn’t want to “interfere.” and.our Original coverage of last week’s crap show interpreted it according to a narrative involving a conflict between boardroom ethics and Altman’s relentless push for commercial OpenAI automation technology. But this story is just a story. We don’t know the specific details that led to Sam’s ouster, but we want to know.
More headlines this week
- Israel uses AI to identify suspected Palestinian extremists. If you’re worried that governments will waste their time weaponizing AI for use in modern warfare, listen to this. According to an article in the Guardian, Israel is currently It uses an AI program called Habsora, or “Gospel,” to identify obvious extremist targets within Palestine. A statement posted on the Israel Defense Forces’ website said the program appears to be used to “generate targets at a rapid pace,” and sources told the Guardian that the program is used by the Israel Defense Forces to He said it helped build a database of 40,000 to 40,000 suspected extremists. The paper reports: “A gospel-like system…[sources said] He played a key role in creating the list of individuals authorized to be assassinated. ”
- Elon Musk spoke out about AI copyright issues this week, and as usual, he sounded ridiculous.. multiple lawsuits They claim that tech companies essentially steal and repackage copyrighted material, allowing them to monetize the work of others, usually writers and visual artists, for free.Elon Musk interjected into this controversial conversation during a speech strange deal book interview this week. Naturally, the thoughts he shared could not be called intelligible. he said and I quote: “I don’t know except that by the time these lawsuits are decided, we’re going to have a digital god. So at that point we can ask a digital god. Hmm. These cases are never decided in a reasonable timeframe.” Beautiful, Elon. You just have to keep an eye on that digital god. Meanwhile, in the real world, legal and regulatory professionals will have to contend with the continued disruption this technology is causing to people far less fortunate than Silicon Valley executives.
- Cruise robotaxis continue to struggle. Cruise, the robotaxi company owned by General Motors, has had a really tough year. The company’s CEO resigned last week amid controversy surrounding the company’s various accidents in San Francisco. this week, reported GM says it will reduce its investment in the company. “We expect the pace of cruise expansion to be more cautious upon resuming operations, resulting in spending in 2024 that is significantly lower than in 2023,” GM CEO Mary Barra said at an investor conference Wednesday. ”, he reportedly said.