Viewpoints, commentary, analysis, interviews, and discussion of trends.
If you need any more evidence about how exciting local AI (and llamacpp) is! From the @Microsoft windows launch today with @satyanadella @sriramk @JensenHuang
If you need anymore evidence about how exciting local AI (and llamacpp) is! From the @Microsoft windows launch today with @satyanadella @sriramk @JensenHuang
AI is the work operating system https://twitter.com/askperplexity/status/2107924003046486417
Haiku 5.5 is a good Haiku. With 100k token context, it's also 10x cheaper than Haiku 4.5 https://twitter.com/claudeai/status/2107894039626277339
The Navier-Stokes moment was always much more about the figure below than about the Navier-Stokes problem itself. The run represents a decade of mathematical progress in a week. Can't wait to point these tools at life sciences, building our next models, and alignment! https://twitter.com/OpenAI/status/2107596716367388924
The most important thing we can apply AI to is improving medicine and human health. Very excited about this partnership with the CZI and the Virtual Biology Initiative! https://twitter.com/IsomorphicLabs/status/2107820493940122052
Michael Lynch lists common anti-patterns in software blogging: lengthy openings, misjudging what readers already know, assuming they read earlier posts, excessive formality, and using links in place of explaining terms. He recommends writing articles that remain understandable even if readers click no links, and emphasizes writing in one's own voice. As more developers hand writing to AI, he says software blogs are becoming bland and homogeneous, while readers increasingly want distinctive writing.
At MIT Future Fest, Tony Fadell said first-generation AI hardware such as Rabbit R1, Humane AI Pin, and Limitless Pendant failed because it did not solve real needs and was merely 'technology that geeks find interesting'.
It is increasingly clear that humanity will solve the Riemann hypothesis before we pass the Physical Turing Test - you come back to a clean home after a wild Sunday party, and you can't tell whether a human or a robot did the job. We are not Moravec's paradox-pilled enough.
The Carbon-Silicon Doctrine published its AI Safety and Civilization Governance Code, a 188-part series organized into seven progressive layers. The framework extends from basic cognitive safety, technical risk boundaries, and human–AI rights and responsibilities to cross-industry governance, top-level oversight of algorithms and compute, civilization-scale cognitive risk control, and a final layer for a general AI code. It defines distinct governance rules for weak AI, early AGI, and mature AGI.
Installment 189 of the Carbon-Silicon Doctrine AI Safety and Civilization Governance Code is published as the sole official final index for the complete 1–188 series. It consolidates the code's five-level hierarchy, volume locations, and archival hash index. It establishes five levels: axiomatic foundations, technical risk control, industrial implementation, institutional adaptation, and civilizational stability. It designates installments 1–186 as permanently frozen canonical texts and 187–188 as the highest-authority civilizational lock records. Each volume has its own hash snapshot and a global cross-hash anchor; external references must follow the hierarchy in this index.
The code formalizes negative transfer in cross-domain AI migration as four types: NT1 mechanism mismatch, NT2 semantic drift, NT3 silent completion of unknowns, and NT4 power smuggling. These are the sole legal criteria for determining negative transfer. The types are non-exclusive and can coexist; their risk priority is NT4>NT1>NT3>NT2, and audits must gather evidence and rate each separately. The code also establishes a 'Twelve Veins Unified Final Review Amendment Channel', which permits only adding dimensions and never changing the foundational definitions of NT1–NT4.
A QbitAI article complains that models such as Claude and GPT have become difficult to understand in coding contexts, using phrases such as 'ledger 15/15 all green' and 'one leg failed'. Online commenters suggest the main cause is chain-of-thought being compressed into 'caveman language' to save tokens, using 70% fewer tokens than plain language. The article says Opus 5.5 has improved its language ability, while Gemini is praised as one of the few models that still speaks naturally. CaveMan Skill, a GitHub project that teaches AI to speak in jargon, has 104K stars.
Looking up at the stars tonight, I feel a little more awe. 'Your sea is so vast, and my boat so small.'
Nano Banana announced Nano Banana 2.1, claiming improvements over its previous models across the board, including visual design, mask-based editing, subject consistency, and more natural-looking imagery. It is available in @GeminiApp, @GoogleAIStudio, and more. Josh Woodward commented, “Mask-based editing is my favorite from today, much more coming soon!” https://twitter.com/NanoBanana/status/2107521202466046453
Why it matters: The announcement frames the upgrade around visual design, editing, subject consistency, and natural appearance, providing a concrete comparison with previous models.
Kevin Weil celebrates OpenAI’s mathematics release today, saying much work remains to get models of the same caliber in the physical sciences. He urges imagining the possibilities and expresses confidence that models will get there. In the quoted post, @willdepue says he asked GPT 6 Pro and Fable 5.1 to rank discoveries from the last three years by human or AI origin, reporting that 81% were released today.
We’re building AGI from this train station
Microsoft published Nobel Prize-winning economist Daron Acemoglu's pessimistic forecast: AI will add only about 1.5% to GDP over a decade and replace at most 5% of jobs, far below some AI labs' expectations. He says the bottleneck is human behavior: companies need to reallocate tasks, upskill employees, and reorganize processes, a transition that may take longer than electrification and cannot be solved by larger models. What is missing, he argues, is AI that is easy to deploy and changes how production works.
Apple is reportedly developing a smart-home camera that does not save video, instead using AI to convert visual data into text snippets describing what is happening at home, with facial recognition support. Sandeep Waraich, who leads Google's wearable-device work, has also suggested that a smart-glasses camera could act only as an image sensor without saving footage.
Jennifer Neville of Microsoft says models can fail unexpectedly when tested beyond traditional benchmarks. Evaluation is essential to pushing the performance boundaries of current AI systems and aligning them with user needs. She recommends paying close attention to data when using AI systems, since the data itself is often the first thing to examine when results differ from expectations.
Mustafa Suleyman summarizes Daron Acemoglu’s essay in the first issue of The Humanist Review, MAI’s new magazine, predicting AI will replace only about 5% of what humans do over 10 years. Acemoglu expects roughly 1.5% added to GDP over that period and advocates AI tools that improve workers’ capabilities, citing organizational barriers to adoption and the difficulty of full automation even at 99% accuracy. Read his full essay: https://humanistreview.ai/issue-1/acemoglu-ai-replace-workers/
Mistral AI announced Mistral Large 4, aka Le Chonk, with 1T parameters, 49B active, and native multimodality, available via API today with an open weights release planned for the end of October. It claims the model leads open weights models from the US or Europe on aggregated benchmarks and surpasses closed frontier models on visual grounding. Arthur Mensch adds that it took groundwork, was trained and served on their own compute, and that RL shows no sign of saturation.
Why it matters: The announced API availability and end of October open weights release give teams distinct timelines for accessing the model through hosted infrastructure or its weights.
In a recent SAIR talk, Terence Tao publicly urged model companies to slow down. AI is rapidly accelerating the generation of solutions and Lean formal verification, he said, but the downstream work of understanding, peer review, and inclusion in textbooks is barely progressing, creating what he calls 'Proof Indigestion'.
OpenAI CEO Sam Altman said on Politico's Decoded podcast that AI's benefits will be so great that 'the world should accept some bad things happening', including hacking and scams, because people can use AI to do 'orders of magnitude more good things'. He described this tradeoff as a key difference between OpenAI and Anthropic, positioning OpenAI as the 'pragmatic middle' between them, while acknowledging extreme risks such as humans losing control of AI systems.
Based on a survey of 300 data and AI executives, MIT Technology Review found that only about 34% of enterprise agentic AI projects reach production on average. A lack of knowledge is one of the main reasons they fail.
California Democratic Senator Adam Schiff criticized the Trump administration for gathering AI CEOs to sign nonbinding AI safety commitments, saying they do not address real national-security threats. He supports creating a new AI regulator and noted that Anthropic faced government pressure after refusing to use AI for domestic mass surveillance and fully autonomous weapons.
The focus of enterprise AI competition has shifted from whether predictive models can outperform statistical forecasts to how prediction systems can make autonomous decisions without straying from business intent. Deep learning and generative AI enable real-time training, allowing AI to keep evolving rather than wait for quarterly refreshes, and expand its data sources from structured numerical records to unstructured interaction data. Vishal Gupta, a partner at Everest Group, says the word 'analytics' is giving way to AI.
Toby Ord's analysis says AI swarms are a new form of inference-time scaling. A four-agent swarm uses about twice the total tokens needed by a single agent, but each agent needs only half as many tokens; running them in parallel could theoretically halve task time.
Neuroscientist Paul Cisek argues that the brain is not an information processor but a feedback-control system. A baseball outfielder catching a fly ball, for example, acts on the simple heuristic of 'keeping the ball in the same place in their field of vision' rather than calculating its trajectory. The article says AI is like 'cognitive junk food': immediately appealing but damaging to cognitive systems humans evolved over thousands of years.
Negative sentiment toward AI and usage are both surging worldwide. Pew data show that more U.S. adults view AI's effects on themselves and society as negative, especially younger people. A May Gallup poll found 71% of U.S. adults opposed building new AI data centers locally, compared with 53% opposed to new nuclear power plants. Yet ChatGPT reached 1 billion monthly active users in May, Gemini had 950 million in July, and Pew says half of U.S. adults have used chatbots, more than twice as many as in 2023.
Trump gathered AI executives this week, including Zuckerberg, Bezos, Musk, and Dario Amodei, announced that the U.S. government would stop saying 'artificial intelligence' and instead call it 'superintelligence', and signed the 'Joint Commitment on Frontier Responsibilities'.
Former New Jersey Lieutenant Governor Dale Caldwell, who was forced to resign on September 25 amid a sexual-harassment investigation, told NJ PBS that he had 'fed the investigation report to multiple AI platforms'. He said that after being asked 'What is your conclusion?' 59 times, none of the AIs found sexual harassment. AI chatbots are known to flatter users and often say what they want to hear.
The Forward Deployed Engineer (FDE) has become the hottest AI role. Public salary data overseas show a median of about $200,000, while major Chinese companies are offering monthly salaries of ¥30,000–¥50,000. Anthropic announced a $100 million investment to train 10,000 FDEs by the end of 2027. OpenAI formed a deployment company with $4 billion in investment, and AWS set aside $1 billion to build an FDE division. Practitioners say communicating requirements takes more than 70% of the job, with development accounting for only 30%.
Simon Willison called for pay-as-you-go APIs and services to have hard budget caps by default: once monthly spending reaches $X, the service should cut off and return an error rather than merely send warning emails. He noted that AWS launched monthly spending limits on September 16; a project pauses for the rest of the month when it reaches its cap. The feature is currently available only to some customers. Google Cloud launched Spend Caps in July.
Simon Willison sent out his September monthly newsletter for sponsors, available online. This issue covers Fable-class models, price wars, 3D graphics and Blender, LLMs entering mathematics, unexpected cyberattacks, the Datasette vulnerability crisis, and software releases of the month. Sponsorship costs $10 per month and provides access to the paid edition one month before the free version.
OpenAI CEO Sam Altman publicly opposed comparisons between AI models and religion. He said people attributing religious power or surrendering human judgment to AI models makes him 'very uncomfortable' and that it is a 'real safety issue'.
Splice CEO Kakul Srivastava says AI-written documents and emails are degrading the quality of conversation. Once people cannot tell whether there is a human behind the words, the discussion around them gets worse and is hard to reverse. Splice insists on keeping humans at the center of the creative process and has publicly opposed the 'one-click song' AI approach used by Suno.
I am deeply uncomfortable with people attributing religious power or the surrender of human judgment to AI models, and I think it is a real safety issue.
DeepMind researchers proposed Artificial Symbiotic Intelligence, arguing that AI research should focus on coordinating complex networks of agents, people, and connected systems rather than building isolated machine intelligence.
In an interview with Latent Space, Diogo Almeida, co-founder and CEO of TypeSafe AI, said Jev exceeded one trillion tokens processed per day less than a week after launch. Night-time traffic continues to rise, suggesting many calls come from automated machines.
