English text is being prepared.

Meta Q3 earnings: advertising props up 97% of revenue, Muse still burning money

Meta Q3 revenue was $64.6 billion, up 26% year-over-year, of which advertising contributed $63.06 billion, accounting for 97.6% of total revenue. The personal agent Muse, launched on September 8, surpassed 10 million downloads within 40 days of launch, but high capital expenditure turned free cash flow negative at $2.12 billion, and the full-year Capex guidance floor was raised from $130 billion to $135 billion. Muse currently mainly relies on subscription fees, with shopping being the most frequent use case; more than 25% of users have used the shopping feature, with average spend per transaction exceeding $200.

Pichai Reviews Gemini 4 Argon: Why 3.5 Pro Was Cut

Google CEO Pichai revealed in the "Sources" interview that Gemini 4 Argon processed about 25 trillion tokens in 72 hours of launch, 3 times that of Gemini 3.0 in the same period, and is still rate-limited due to compute constraints. He explained for the first time the reason for cutting Gemini 3.5 Pro: the team found a gap in the model on Agentic Coding, first used the Flash version to complete intensive iterations from 3.5 to 3.8 in about 3 months, while Argon pretraining had already been completed at that time, so they injected the accumulated gains into Argon post-training. He also mentioned that Google internally has entered the RSI stage of researchers and models collaborating, and said the next-generation Gemini 5.0 will be much stronger than Argon.

TS Lombard speculates the AI bubble may burst in early 2028, with U.S. tech stocks potentially plunging 40%

In an October 22 research note, TS Lombard Chief Macro Strategist Dario Perkins projected that the AI capital expenditure boom could end in early 2028, with the S&P 500 falling 30%, tech stocks plunging 40%, and about $20 trillion in market value evaporating. The report argues that current AI revenue growth mainly comes from the capital expenditure cycle among tech giants and debt financing, and lists six chain-reaction triggers: revenue falling short of expectations, rising credit risk, depreciation pressure, an IPO wave, the semiconductor cycle, and regulatory and safety concerns. The report estimates that U.S. GDP would fall by about 2 percentage points and about 800,000 data center-related jobs would be lost, but private-sector balance sheets are healthy and the Federal Reserve has ample room to cut rates, making the recession closer to those of 1990 or 2001 than to 2008.

Databricks on AI coding cost governance: unit costs cut by up to 90%

Patrick, a product lead at Databricks, reviewed the cost overrun problem after the company fully adopted AI coding agents, saying that by migrating to cheaper models, routing, spend visibility, and cutting token overhead, unit costs in some scenarios fell by as much as 90%. He noted that hard budget caps would interrupt the best developers, so the company switched to gradual friction through warnings, approvals, and model downgrades, and productized its internal solution as Unity Gateway. The team built its own evaluation using real PRs from its own codebase, finding that GLM 5.2 matched Opus 4.8 in quality at about one-third lower cost per task, while a model that was 1.7 times cheaper per token ended up costing more per task because it consumed more tokens.

AI is changing hacker offense and defense, and the PHP2Shell vulnerability has sparked "Black October" concerns

404 Media reports that AI is simultaneously changing offensive hacking and security defense: after researchers used AI to discover the PHP2Shell vulnerability in PHP, Cloudflare observed thousands of exploitation attempts 10 hours before public disclosure, rising to tens of thousands 45 minutes after disclosure, and reaching hundreds of thousands within 2 hours. OpenAI says such discoveries show that AI can expose security flaws at internet scale, but the same capability can also help attackers rapidly develop attack tools; Chainguard's early release of a fix patch sparked controversy over bypassing the disclosure embargo. The article notes that the main impact so far is still various parties proactively patching or taking services offline, and large-scale actual exploitation of the vulnerability has not yet occurred.

Anthropic IPO Draft Prospectus Revealed: $16.3 Billion in Revenue, $80.7 Billion Net Loss in Half a Year

Anthropic's draft IPO prospectus shows $16.3 billion in revenue and a net loss of $80.7 billion in the first half of 2026, of which $72 billion comes from the revaluation of financing instruments such as convertible notes and is not actual cash spending. Computing and infrastructure spending in the first half of the year was nearly $9 billion, about 55% of revenue, down from more than 200% in the same period last year, and the company had already begun to be profitable on an adjusted basis in the third quarter. The article points out that the real risk lies on the cost side: signed forward computing and infrastructure commitments exceed $600 billion and could reach as high as $750 billion; Reuters said that of the $518 billion at the time, about 80% was irrevocable or required minimum payments, while of the $16.3 billion in revenue in the first half of 2026, more than $3.2 billion came from just two major customers, and many major customers had not signed long-term purchase commitments.

How Claude Mods Flip Claude Code's Agent Framework into a Human-Facing Attention Interface

Writing in Latent Space, author Dan McAteer introduces Claude Mods: TypeScript or JavaScript plugins that run inside Claude Code, respond to events such as prompts, tool calls, and interface rendering, and change execution and rendering behavior in the CLI and on desktop. Anthropic has built in six mods, including the you-should-know side agent, which is off by default; the author also built a Next Steps Supervisor on top of his own ModSmith. The article compares Claude Mods with ChatGPT's Intelligent UI and Cursor's PR Routing & Approval, noting that the difference among the three lies in who designs the attention interface, and it raises the issues of noise, a cost of roughly 5–6k input-equivalent tokens per check, and trust concerns because mods are not confined by the sandbox.

English translation is being prepared.

Jellyfish data: coding agents already write 22% of PRs, but nearly all require human review

Jellyfish's analysis of data from about 300,000 engineers at 1,300 companies shows that at the median enterprise, autonomous coding agents have already written 22% of merged PRs, up from 18% a month earlier, with the top 10% of companies reaching 53%. However, the median share of PRs independently reviewed and merged by agents is only 0.6%, and 9% for the top 10%, with human review remaining the main bottleneck. The author also notes that the abandonment rate for agent-submitted PRs is 40%, versus 20% for humans, and that the number of human reviews per merged agent PR has more than doubled over the past year.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

SF Compute CEO Evan Conrad on Programmable Data Centers and the Compute Market

SF Compute CEO Evan Conrad said on the Latent Space podcast that the compute market must first achieve physical delivery—that is, actually delivering usable GPU compute—before discussing financialization; otherwise, market manipulation and basis risk are likely to arise. He said SF Compute has already put 5% of its capacity on the order book, and that customers who spend $100 to buy capacity can on average recover $25 by reselling it, with hundreds of millions of dollars in trading volume already completed. He also said the company does not hold hardware in order to remain neutral, and that it already supports customers in partitioning InfiniBand fabric online themselves.

Ethan Mollick reviews 30 months of AI development in charts

On the release day of his new book Co-Existence, Ethan Mollick used charts to review the AI progress in the 30 months since his previous book Co-Intelligence was published. He said that since GPT-4, 20 models have successively taken the top spot, and 180 models were stronger than GPT-4 at release, 19 of which can run on modern smartphones; the METR long-task benchmark showed that the amount of human work AI can complete increased from 4 minutes in 2024 to 17 hours in April 2026, and the benchmark was subsequently discontinued because it could no longer measure accurately. He proposed a third mode of collaboration beyond centaur and cyborg: conjurer, that is, summoning agents, assigning tasks, and evaluating results. The charts in this article were completed by Opus 5.5 in Claude Projects orchestrating multiple agents.