Oracle is reportedly preparing more layoffs as AI-driven cloud expansion pushes free cash flow negative. Business Insider reports some parts of the company could lose double-digit percentages of staff, with payroll reductions targeted before September 1. The plans follow a drop from 162,000 to 141,000 employees. Oracle spent $55.7B on capital expenditures against $32.0B of operating cash flow, leaving free cash flow at negative $23.7 B. It raised $43B in debt and $5B in equity during fiscal 2026, and expects roughly $40B more financing in fiscal 2027. However, that capex burden accompanies rapid growth, with total revenue rising 17% and cloud infrastructure revenue climbing 77%.

Grok 4.6 gets 98.4% of Fable 5's Intelligence Index score at a fraction of its API price. 5X cheaper on input tokens and 8X less for output tokens vs Fable 5. Looks clearly the strongest intelligence-per-dollar equation. Grok 4.6 scores 61 versus Claude Fable 5 Max's 62, just a 1-point or 1.6% relative gap on that composite score, while Grok costs 80% less on input tokens ($2 vs $10/M) and 88% less on output tokens ($6 vs $50/M).

Grok 4.6 just dropped. Clearest strength is professional agent work, with leading results against GPT Sol Max and Fable 5 Max on GDPVal-AA v2 and AA-Briefcase. - matches GPT-5.6 Sol Max at 61 on Artificial Analysis while charging $2/$6 per million input/output tokens. - 1753 on GDPVal-AA v2 and 1577 on AA-Briefcase, both ahead of Sol Max and Fable 5 Max. Those benchmarks measure real-world agent tasks and agentic knowledge work, making them closer to research, analysis and multi-file deliverables than isolated question answering. On coding, 69.9% on CursorBench v3.2 beats Sol's 67.2%, while DeepSWE and Terminal-Bench leave Grok behind both Sol and Fable. SpaceXAI attributes the jump to a longer training run, regenerated SFT trajectories, model-based trace filtering, and agentic RL across coding, web development, CAD and kernel optimization. It also reports more self-testing on long trajectories, with the model checking its work before continuing, directly targeting error accumulation across multi-step agents.

OpenAI finally launched a ChatGPT desktop app for Linux. ChatGPT, Work, and Codex into one interface. Codex already worked on Linux through its CLI, so the new app mainly adds the integrated desktop layer. The preview supports Ubuntu 24.04 and 26.04 LTS, Debian 13, and Fedora 43 and 44 on x64 and ARM64.

New Google paper shows LLM infrastructure optimization does not have to be a giant brute-force search if an agent can understand why the system is slow. The key shift is from “search everything” to “understand the bottleneck, then search the tiny part that matters.” Instead of asking, “Which of these 100 configurations is fastest?”, this work asks, “What is actually causing the slowdown?” That one change dramatically shrinks the search. Its Analyzer Agent reads profiler traces and classifies the bottleneck as compute, memory, or communication, then a Proposal Agent retrieves engineering knowledge and generates three targeted ways to map data, model, and sequence parallelism across TPU chips. Across eight production workloads spanning dense and MoE models, pre-training, post-training, serving, and TPU systems from 2 to 2,048 chips, the human-validated production configuration appeared in PROMPTS' first batch every time. In seven of eight cases, it was also the first configuration tested, and the agent's top-ranked proposal matched what engineers ultimately adopted in 87.5% of cases.

Oracle is reportedly preparing more layoffs as AI-driven cloud expansion pushes free cash flow negative. Business Insider reports some parts of the company could lose double-digit percentages of staff, with payroll reductions targeted before September 1. The plans follow a drop from 162,000 to 141,000 employees. Oracle spent $55.7B on capital expenditures against $32.0B of operating cash flow, leaving free cash flow at negative $23.7 B. It raised $43B in debt and $5B in equity during fiscal 2026, and expects roughly $40B more financing in fiscal 2027. However, that capex burden accompanies rapid growth, with total revenue rising 17% and cloud infrastructure revenue climbing 77%.

OpenAI finally launched a ChatGPT desktop app for Linux. ChatGPT, Work, and Codex into one interface. Codex already worked on Linux through its CLI, so the new app mainly adds the integrated desktop layer. The preview supports Ubuntu 24.04 and 26.04 LTS, Debian 13, and Fedora 43 and 44 on x64 and ARM64.

New Google paper shows LLM infrastructure optimization does not have to be a giant brute-force search if an agent can understand why the system is slow. The key shift is from “search everything” to “understand the bottleneck, then search the tiny part that matters.” Instead of asking, “Which of these 100 configurations is fastest?”, this work asks, “What is actually causing the slowdown?” That one change dramatically shrinks the search. Its Analyzer Agent reads profiler traces and classifies the bottleneck as compute, memory, or communication, then a Proposal Agent retrieves engineering knowledge and generates three targeted ways to map data, model, and sequence parallelism across TPU chips. Across eight production workloads spanning dense and MoE models, pre-training, post-training, serving, and TPU systems from 2 to 2,048 chips, the human-validated production configuration appeared in PROMPTS' first batch every time. In seven of eight cases, it was also the first configuration tested, and the agent's top-ranked proposal matched what engineers ultimately adopted in 87.5% of cases.

Memory pricing now sets the dating value of Samsung and SK Hynix chip engineers. This is happening after large bonuses are expected to these employees. Averages about $400K/$500K per employee at Samsung and SK Hynix.

Wow, so many hard problems in a single sequence here. And each one is a reason machines have not yet replaced human hands on assembly lines. @DynaRobotics shows their dual-arm robot cracking a fortune cookie and extracting a tiny paper slip Deformable objects are one of the toughest problems for robotics to handle. The hardest part may actually start after the cookie cracks. One rigid object suddenly becomes several fragments plus a thin, partly occluded piece of paper, so the robot has to continue the task after its entire manipulation scene has changed. The policy has to update its internal stress model on the fly as the cookie opens, something that exposes the limits of most current foundation models on irreversible state transitions.

FT: ByteDance is reportedly pre-training an AI model with up to 10T parameters, far exceeding Kimi K3’s 2.8Trn param size. FT also reported that ByteDance has avoided distilling rival models for more than a year, preferring independent model development. If the reported run succeeds, ByteDance will have shown it can execute frontier-scale pretraining without leaning on a rival model as teacher.

Seedance 2.5 landing in Arcads, makes the workflow more interesting. Arcads begins with formats already performing in your category, uses Claude to turn them into creative directions, and now brings Seedance 2.5 into the production layer. Now the same system can move closer from research to finished video. The model will have unlimited access for the first 24 hours

New Google Paper says classical game theory predicts betrayal, but similar AI agents can rationally choose cooperation because their decisions are predictably linked. The big claim is that similar AI agents can rationally cooperate even when they cannot communicate or benefit later, because each agent’s own planned choice helps it predict what the similar agent will choose. Classical game theory treats each player as separate, so it predicts defection in a final one-shot Prisoner’s Dilemma. The authors instead model an AI agent as part of the world it predicts, including uncertainty about its own behavior. When past choices suggest another agent thinks similarly, considering cooperation makes that partner’s cooperation seem more likely too. Gemini and Gemma agents played varied games before a final dilemma, either directly or through shared third-party encounters. With enough evidence, identical agents cooperated strongly, related models cooperated less, and random opponents usually faced defection. The proposed embedded equilibrium may better predict AI societies, while warning that similar AIs could favor each other over humans. – arxiv. org/abs/2608.03958 Title: "A game theory for foundation models shows new paths to rational cooperation through similarity inference"

AI is Jevons paradox territory. After GPT-5.6 Luna's token price fell by 10x, token consumption jumped by more than 10x. The real AI demand shock has barely started. Only 17.8% of the world’s working-age population was using generative AI in Q1 2026. (For Microsoft Research Report) Jevons will get much more interesting when the other 82% shows up.

Muse Spark 1.2 puts Meta on Artificial Analysis's Pareto frontier at $0.40 per task. i.e. on that frontier, no model in the comparison is simultaneously cheaper per task and higher on the Intelligence Index. Muse Spark 1.2 also reaches roughly Claude Opus 4.8-level intelligence at about one-fifth of that model's $2.03 task cost. It scores 6 Index points below Claude Opus 5 while costing roughly one-sixth as much. The metric goes beyond just the list price because Artificial Analysis weights the input, cache, reasoning, and answer tokens actually consumed across its nine Index evaluations. This is so significant because for long-running agents, small cost differences compound across model calls, reasoning tokens, tool steps, and retries.

The Philippines runs a $40B outsourcing industry employing 1.9M people, and AI is now taking the jobs it was built on. BBC published a piece. That trade was built in the early 2000s on one advantage, because the country could supply English-speaking staff more cheaply than India. Governments handed out tax breaks and infrastructure, so Accenture, Concentrix and Teleperformance filled Manila with campuses holding hundreds of thousands of workers. That worked while buyers wanted a person at a desk for less. Now those same clients expect suppliers to run automation as well, which changes what a headcount contract is worth. More than two-thirds of the industry association's members are already running AI pilots, and its president concedes a layer of jobs has gone. He says most of them moved into other roles. The mechanics are less reassuring, since the first stage of automation shows up as extra work rather than less. --- bbc .com/news/articles/cgr7nxve05go

"OpenAI and Anthropic are pretty different animals from a capital efficiency perspective. And Anthropic clearly has a dramatically lower cost per token than OpenAI. They burned maybe 80% less than OpenAI. So as businesses, they clearly have very different structural ROICs." - A solid, insightful interview featuring Gavin Baker (@GavinSBaker ), U.S. hedge fund manager and Chief Investment Officer of Atreides Management. ---- From "Invest Like The Best" (@patrick_oshag ) YouTube channel, (link in comment)