The Kobeissi Letter is a market commentary account. The S&P 500 Excluding AI is an index view that leaves out information technology and communication services stocks, where most large AI companies sit. The Nasdaq 100 is heavy in large technology stocks.

The Kobeissi Letter posted on Saturday that what it calls a stealth bear market is in full swing. According to the post, the Nasdaq 100 erased 4 trillion dollars in 25 days, while the S&P 500 Excluding AI hit a record high on Friday.

The post says AI stocks fell over 13% in October while the S&P 500 Excluding AI gained 8%, and it opens a thread to explain why. Its attached chart, which tracks the period from early October, shows the S&P 500 excluding information technology and communication services up 4.79% and the Nasdaq 100 down 9.18%, and labels the move the largest monthly drop on record for AI stocks. The figures in the text and the chart cover slightly different measures and periods, and they come from The Kobeissi Letter rather than an official index provider.

The post lands after a week of news that tested confidence in AI spending, including reports in this deck's Oct 30 edition that OpenAI cut the size of its fundraise and that S&P lowered Meta's outlook over heavy AI spending. The post does not argue that AI demand is falling, only that investors moved money out of AI names during the month.

Watch the first trading days of November to see if the gap between AI stocks and the rest of the market keeps widening or reverses. Upcoming earnings from chip makers and cloud providers will test whether AI spending plans hold. Also watch whether AI startups find it harder to raise money at high valuations if the selloff lasts, and whether the drop changes how much big tech plans to spend on data centers next year.

In recent weeks OpenAI has released more than 700 AI generated proofs, including a Navier Stokes result, and many mathematicians reacted with alarm. Sam Altman discussed the backlash on the Cheeky Pint podcast with Stripe cofounder John Collison, released on Thursday, Oct 29. Polymarket posted the quote on Saturday, and it spread widely.

On the Cheeky Pint podcast with John Collison, Sam Altman said OpenAI "did let it get out of control for one week on the math," and that this was intentional. According to the published transcript, he said the company had to push hard to learn how far teams of AI agents could go, because when it did the Navier Stokes proof it understood almost nothing about how to scale them: which model to use, how many copies, and how long to let them run.

Altman put the cost of the Navier Stokes work at around $10 million, "maybe a little more," and said later results cost far less on average as the team learned. Asked whether the math effort ran to hundreds of millions of dollars, he agreed, adding that the figure covers much more discovery than OpenAI has published. He confirmed OpenAI holds more than the roughly 700 proofs it has released, and said he bets the company could redo Navier Stokes at one hundredth of the cost a year after it was done.

Collison noted that mathematicians were up in arms, and Altman agreed they were not happy. He said he feels "unequivocal" that humanity deserves fundamental knowledge and that OpenAI should not sit on major discoveries, while saying the company owes it to people in those fields to find a way for them to keep a fulfilling role. He described a weekend when staff threw their hardest research problems at a new model, and said watching it solve them brought him both excitement and sadness.

Polymarket's post framed the remark as an admission amid backlash. The full transcript shows Altman presenting the week as a deliberate experiment rather than a mistake.

Watch whether OpenAI changes how it releases math results, for example with fuller papers, outside review or formal proofs checked in Lean before publication. Mathematicians and advisory groups may press for clearer credit and verification rules. Altman suggested more unpublished results exist, so further releases are likely. Also watch the cost claims: if similar proofs become far cheaper, other labs may run their own large agent math efforts.

Terafab is the name Musk has used for a very large chip factory he wants built to supply AI compute. A mass driver is an electromagnetic launcher that throws payloads using electricity instead of rocket fuel. X Freeze is a fan account that often posts clips of Musk.

X Freeze posted a clip on Saturday in which Elon Musk answers what he called the most ambitious question: what comes after a Terafab. The post sums up his answer as a petawatt scale future built on the Moon.

According to the post, the plan is to use Tesla Optimus robots, humans and local lunar manufacturing to build huge compute infrastructure, then launch it into deep space with electromagnetic mass drivers. Because the Moon has no atmosphere and about one sixth of Earth's gravity, the post says such launchers could reach escape velocity on electricity alone, with no rocket fuel burned on the lunar surface. It frames this as scaling compute by three more orders of magnitude, from terawatt to petawatt, a 1,000 times increase.

The X Freeze post quotes a reply Musk wrote the same morning, saying that launching super intelligence satellites into deep space with a mass driver on the Moon can achieve more than a petawatt per year of capability, which he said would far exceed the best case value of Earth's economy. Musk then reposted the clip. The video itself was first shared in an earlier X Freeze post, so the remarks are not new today, but the idea drew fresh attention.

This is a long range vision, not an announced project, and Musk gave no timeline or cost.

Watch whether Tesla or SpaceX attach any concrete plan, budget or timeline to lunar manufacturing, since today this is a vision rather than a program. Progress on the Terafab chip factory itself is the nearer signal to follow. Engineers and skeptics will likely test the physics and economics of mass drivers and space compute. Any Optimus milestones would also show whether robots could do this kind of work.

Aaron Saunders posted this as a member of the Google DeepMind team. CAD, or computer aided design, is the software engineers normally use to draw parts. Degrees of freedom count the independent ways a robot can move.

Aaron Saunders posted a Halloween video on Saturday. He says that while most jack o lanterns just sit on porches, the Google DeepMind team let AI agents design, simulate and bring one to life.

According to the post, the team built a physical dancing robot with 12 degrees of freedom directly from a single plain English prompt, with no manual CAD required. The 33 second video shows a robot shaped like a carved pumpkin.

The post does not name the agents or models used, how long the process took, or how much human help went into assembly, and Google DeepMind has not published a paper or blog post about it that the post links to. The X account posting it is new, so the claims rest on the post itself.

Even so, the idea fits a trend of AI agents moving from writing software to designing physical objects, where the hard parts are getting parts that fit, motors that work together and a design that can actually be built.

Watch for a DeepMind paper, blog post or demo that explains the system behind the robot, which would show how much of the work the agents did on their own. Also watch whether similar agent driven hardware design shows up in other labs. A public tool would let engineers test whether this works beyond a holiday demo, and whether it can handle larger and more useful machines.

Anthropic trains Claude against a written set of principles it calls a constitution. The Associated Press reported on Oct 25 that AI companies are turning to religious thinkers for guidance, and unusual_whales shared that report on Saturday evening, where it drew wide attention. Scientific American has reported that about 15 religious thinkers met Anthropic in late March to discuss Claude.

According to the post, which cites the Associated Press, Anthropic says it has met with scholars, clergy, philosophers and ethicists from more than 15 religious and cross cultural groups to help shape how Claude handles moral questions.

The AP report, published on Oct 25, quotes Anthropic cofounder Chris Olah speaking earlier this year at the Vatican presentation of Pope Leo XIV's encyclical on AI: "These are not questions a lab can answer. They are questions traditions like yours have carried for millennia." The AP headline notes that experts say there is more to religion than rules.

Scientific American reports that around 15 religious thinkers met Anthropic in late March. Attendees said the aim was not to make Claude pious but to refine its constitution, the principles the company uses to train the model to critique and revise its own answers. One participant said Anthropic wants to shape a disposition toward good behavior rather than write a rule for every case.

A separate AP report covers a roundtable where tech companies sought guidance from faith leaders on AI. The story is not new this weekend, but the repost drew more than 150,000 views and renewed debate about who decides the values built into chatbots.

Watch whether Anthropic publishes more detail on which groups it consulted and what changed in Claude's constitution as a result. Other labs, including OpenAI, may describe similar outreach. Critics will likely keep asking how labs reconcile conflicting traditions and who is accountable for the final choices. Also watch for regulators or religious bodies responding, especially after the Vatican's recent encyclical on AI.

Optimus is the humanoid robot Tesla is developing. Tesla Owners Silicon Valley is a fan account that often posts clips of Musk.

Tesla Owners Silicon Valley posted a clip late on Friday of Elon Musk explaining why the hand is the hardest part of building Optimus. Musk reposted it on Saturday.

In the quoted remarks, Musk says that building a humanoid robot has made him think a lot about how the human body works, and more than anything about how incredible human hands are. He notes that almost all the muscles that control the hand are actually in the forearm, pulling the fingers like puppets.

That design point matters for robot builders. Putting the motors that drive the fingers in the forearm, the way the body does, keeps the hand itself light and compact, but it means routing tendons or cables through the wrist. Hands with many independent joints are widely seen as one of the toughest parts of humanoid robots, because useful work like picking up tools or handling small parts depends on fine control and touch.

The post does not say when or where Musk made the remarks, and it gives no new specifications or dates for Optimus. The clip is commentary rather than a product announcement, but it drew wide attention after Musk shared it.

Watch for any new Optimus demonstration that shows the hand doing delicate tasks, which would back up Musk's focus on it. Tesla may share hand specifications, such as the number of moving joints, at a future event or earnings call. Rival humanoid makers are working on the same problem, so comparisons of hand dexterity will be a useful way to judge progress across the field.

Cleo Abram hosts Huge If True, a video series about science and technology. Neuralink is Elon Musk's brain computer interface company, and R1 is the robot it uses to implant its device.

Cleo Abram posted the full Huge If True episode on Neuralink and the future of brain computer interfaces on Saturday. She says she toured Neuralink and got a close up view of its new R1 surgical robot, to see the future the company and Elon Musk are building and how it is already changing lives.

The video runs about 18 minutes, and the chapters listed in the post walk through what a brain chip is, how the brain works, whether a brain computer interface can read thoughts, how the chip gets into the brain, what is inside the implant, how it is built, how a robot does brain surgery, and how such devices are already used today.

Elon Musk quoted the post a couple of hours later with the words Neuralink deep dive, and his quote drew more than a million views.

The episode is a feature rather than a company announcement, and the post does not list new trial results, patient numbers or product dates. It is included here because it drew wide attention and offers an accessible look at the R1 robot.

Watch for Neuralink updates on patient numbers and trial results, which would show how fast the R1 robot and the implant are being used. Rival brain computer interface companies are also running trials, so comparisons will grow. Also watch whether regulators set a path toward wider approval of these devices, and whether Neuralink shares more about how the R1 robot performs in surgery.

Lumina is an X account that posts about AI models and benchmarks. GLM is the family of large language models from the Chinese lab Zhipu AI, also known as Z.ai. The post refers to GLM 5.3 as the most recent release.

Lumina posted on Saturday that GLM 5.5 is well into post training. This is an unverified claim. Zhipu AI has not announced GLM 5.5, and the post does not say where the information comes from.

According to the post, Zhipu seems to have been training internal models of 32B, 122B and 256B parameters that will be merged into the next generation base model, which Lumina says will be GLM 5.5. Lumina adds that it has been a couple of months since GLM 5.3 and says it hopes the new model is released relatively soon.

The attached screenshot shows a highlighted line of text that describes a next generation GLM base model called GLM 5.5, coding post training end to end, quant finance training environments and a coding sandbox, post trained internal 32B, 122B and 256B models merging into the base model, and a research topic on self evolution in financial coding. The post does not identify where the screenshot was taken from, so its origin cannot be confirmed.

The post drew more than 100,000 views, showing strong interest in what Zhipu releases next. Until Zhipu confirms details, the model sizes, the training plan and any timing should be treated as Lumina's claims.

Watch for any official word from Zhipu AI or Z.ai about GLM 5.5, which would confirm or rule out this report. If the model ships, look for its size, its license, whether the weights are open, and how it scores on coding benchmarks against other open models. Also watch whether the finance and coding focus described in the screenshot shows up in the release.

Bill Ackman is a hedge fund manager who runs Pershing Square. Fireside Alpha is an X account that posts clips of investors and executives. Neoclouds are newer cloud companies that rent out GPU capacity for AI, often funded with debt.

Fireside Alpha posted a 71 second clip on Saturday of Bill Ackman speaking at a Milken Institute event. The post quotes Ackman at length.

In the quoted remarks, Ackman says he thinks AI demand is sustainable because intelligence is an enormously valuable asset, so demand for the compute behind it will last. He says the hyperscalers building much of the infrastructure have the financial resources and creditworthiness to access capital.

Ackman says he worries more about the pure play neocloud companies, which he describes as a bit like highly levered real estate developers playing a theme. He says they are much more at risk and much more sensitive to the cost of capital.

He then points to SpaceX, saying Elon Musk gave projections on an earnings call for how many gigawatts of data centers he plans to open, and that if Musk meets his objective he might have to borrow a trillion dollars of capital. Ackman says companies are now competing at sovereign levels for debt, and that a lot of demand for capital against a finite supply makes rates persistently high.

The trillion dollar figure is Ackman's own estimate, not a number from SpaceX. The post quotes an earlier Fireside Alpha clip of Musk explaining how SpaceX casts turbine parts to bring power online faster.

Watch whether SpaceX or Musk's companies announce large debt raises tied to data centers, which would test Ackman's estimate. Also watch borrowing costs for neocloud companies and whether any of them struggle to refinance if rates stay high. Further comments from other investors on AI debt, and moves in bond markets, will show whether this concern spreads. Earnings calls from hyperscalers may also reveal how they plan to fund their own buildouts.

Bill Ackman is a hedge fund manager who runs Pershing Square. Brain computer interfaces read signals from the brain to control devices or restore lost functions, and Neuralink is the best known company in the field. This deck's story on Cleo Abram's Neuralink episode covers how the Neuralink implant works.

Fireside Alpha posted a 76 second interview clip of Bill Ackman on Saturday. The post quotes Ackman speaking about his daughter and about brain computer interfaces.

In the quoted remarks, Ackman says his daughter lost basic functions including the ability to walk, speak and see, and is now regaining function, relearning how to speak and walk. He says the biggest challenge is that optic nerve damage from pressure on the brain has caused her to lose probably 98% of her vision.

Ackman says he is optimistic even there. He says he has spent a lot of time in recent months meeting brain computer interface companies, with Neuralink being the most famous, along with other promising lower intervention technologies.

He says AI will make it possible to get more signal from the brain and interpret it much more effectively than before, and that the blind will see again and paraplegics will walk again. He says it is a matter of years rather than months, but not too many years.

These are Ackman's predictions, not results from any company. The clip comes on the same day Cleo Abram released her Neuralink episode, which is covered separately in this deck.

Watch for trial results from Neuralink and other brain computer interface companies on restoring vision, which would test Ackman's timeline. Also watch whether Ackman or Pershing Square invests in any of the companies he says he met. Regulatory steps toward wider approval of these devices, and progress on AI decoding of brain signals, will show how close his prediction is in practice.

Theo runs the t3.gg developer channel and is a well known voice among web developers. Opus 5.5 is a Claude model from Anthropic. Effort settings let users choose how much reasoning the model does before it answers, and higher effort usually costs more and takes longer.

Theo posted on Saturday that someone he calls Julius convinced him to move from extra high to high effort on Opus 5.5 a week or two ago. He says Julius was right, and that the high setting is better almost all of the time.

The post is a personal usage tip rather than a product announcement or a benchmark. It does not say which tasks Theo tested, how he judged the results, or whether the change also saved time or tokens.

The post still drew more than 280,000 views and over 300 replies, which shows how many developers are tuning effort settings to get the most from Anthropic's models. Effort settings trade speed and cost against how long a model thinks, and users often assume the highest setting gives the best answers.

Theo's report suggests that more reasoning is not always better for everyday work. It is one developer's experience, and results may differ by task, so readers should treat it as an opinion to test rather than a settled finding.

Watch whether Anthropic publishes guidance on when to use each effort setting for Opus 5.5, or changes the defaults. Other developers may share their own comparisons, which would show whether Theo's finding holds across coding and writing tasks. Also watch for benchmark results that compare effort levels on cost, speed and quality, and whether other labs see the same pattern with their own reasoning settings.

Steven Mark Ryan is a commentator who posts often about Tesla and Elon Musk. Musk reposted his prediction on Saturday night, and the repost drew far more views than the original. The post comes as AI stocks had their worst month on record in October, according to The Kobeissi Letter.

Steven Mark Ryan wrote that superintelligence will "eat many entire businesses, industries and professions in the next few months and quarters," that the rest will follow, and that the physical world will take longer because it needs a lot of humanoid robots.

Elon Musk reposted it without comment late on Saturday. The X API counts more than 400,000 views on the repost and about 250,000 on the original, with over 200 replies.

The post gives no evidence, names no industries and does not define superintelligence. It is a forecast, and readers should treat it as opinion. It does line up with views Musk has shared before, including his push for Optimus humanoid robots and for huge amounts of compute on Earth and in space.

The timing stands out. It came at the end of a month in which AI stocks fell sharply, OpenAI cut its funding target, and several market commentators warned of an AI bubble. Bold claims about near term disruption from AI leaders and their supporters keep attracting large audiences even as investors grow more cautious.

Watch whether Musk or xAI back the claim with product news, such as the Grok 4.8 release Musk has teased, or with Optimus milestones. Labor data and company layoff announcements over the coming quarters will show whether AI is replacing whole professions as fast as the post predicts. Expect continued debate between those who see rapid disruption and those who see a bubble.