Two hundred and fifty years after the signing of the Declaration of Independence, a new commercial from Google asks: What if the Founding Fathers had access to Google Workspace?
With the tagline “Group project, but make it 1776,” the ad depicts a largely unseen Thomas Jefferson mid-draft when he gets a nagging text from Ben Franklin, leading to a very Google-centric collaboration process. Edits are suggested in Google Docs, a meeting gets scheduled in Google Calendar and conducted remotely via Google Meet (with every single attendee apparently turning their camera off?), then the whole thing is finalized with e-signatures; cue the fireworks. — Read More
Daily Archives: July 7, 2026
A global workspace in language models
As you read this sentence, circuits in your brain are adjusting your posture, controlling your breathing, and transforming lines and curves on the screen into recognizable words. Most of this processing is invisible to you. But some of what takes place in your brain you do have access to—an image that pops into your head, or a deliberate plan you make about where to go shopping. Neuroscientists and philosophers sometimes refer to the latter type of brain activity as “consciously accessible,” to distinguish it from all the other processing that goes on unconsciously. This activity has special properties: we can describe it, control it, and use it for deliberate reasoning, in contrast to all the automatic processing that goes on without our awareness.
In a new paper, we present evidence that a similar distinction has emerged in modern language models like Claude. We find that Claude has developed a small collection of internal neural patterns that, compared to all its other internal processing, play a special role. — Read More
A Forlorn Hope of Fortran Modernisation
For decades, people in IT had taken delight in drafting Fortran’s obituary. Yet, this old language lives on. But in recent years, the Fortran user community has begun sounding alarms: Fortran shops are having difficulty finding young programmers to replace those who are leaving the workforce, because the young are not willing to devote their careers to this archaic language. At present, no language can rival, let alone surpass, Fortran when it comes to implementing long-lived, large-scale, massively-parallel scientific and engineering applications; not even C and C++. Yet, modern programmers know nothing about Fortran, nor have they any interest in it. Suffice it to say, Fortran has an image problem.
… It would seem that trying to shore up this mid-century language for the grind of the 21st Century verges on insanity. Not so. I contend that Fortran modernisation is worthwhile and necessary. I admit, though, that refreshing Fortran for the 2020s is but a forlorn hope, at present. — Read More
The Realities of AI Video Surveillance
The Financial Times has a good article on how AI is changing the capabilities of video surveillance, with information from both Israel/Iran and Russia.
I wrote about this sort of thing a few years ago, how AI enables mass spying in the way that computers and networks enabled mass surveillance. The interesting development in the article is that AI allows people to ask natural language questions about video footage to AIs—and AIs can answer them. — Read More
The Robots Are Here
Robots are here, and they’re going to change the world, with Unitree currently in pole position. Joining me to discuss are Niko Ciminelli, longtime SemiAnalysis advisor, robot kid, venture investor, and Western New York native — go Bills! — along with Reyk Knuhtsen, robotics lead at SemiAnalysis. Lily Ottinger is co-hosting.
Our conversation covers:
Why robots are the real general-purpose technology
How everyone keeps underestimating Unitree
China’s edge
Why you can’t AI your way out
What America should do next
Read More
The AI Superforecasters Are Here
The annual prediction market conference was earlier this month. This was the year prediction markets went from an obscure hobby to a multi-billion dollar industry; from semi-illegal to having the President’s son as an advisor. I can’t remember if anyone talked about any of that. It didn’t even register. All eyes were on the AI superforecasters.
The extending-lines-on-graphs community has long predicted that AIs would beat the best human forecasters sometime in 2026 – 2027. What did you expect the bots-finally-beat-humans-at-predicting-the-future moment to look like? Vibes? Papers? Essays? In retrospect, sure: it will look like AIs making crazy profits on prediction markets and beating the stock market by some comfortable amount.
But what happens next? — Read More
The context engineering playbook (Claire Gouze)
Context is everything in data right now. Everyone is looking for the same thing: a single place where you can ask a natural-language question and get back a reliable answer. With that, you can build conversational analytics, but you can also build essentially any agent that needs to connect to your organization’s data. As long as you have context.
… Lots of people are talking about building context layers and hiring context engineers. [nao co-founder and CEO Claire Gouze] and her team are doing it: they’ve authored a context engineering playbook with specific guidance on how to build your own context layer and how to create evals. They’ve built a community learning together in the open, and they’ve built tooling to make it easier. What I appreciated most about this conversation was its pragmatism. Rather than talking about context, we should all get to work engineering it. Claire’s playbook is a great place to start. — Read More
Better Models: Worse Tools
A very strange Pi issue sent me down a rabbit hole over the last two days. The short version is that newer Claude models sometimes call Pi’s edit tool with extra, invented fields in the nested edits[] array. And not Haiku or some small model: Opus 4.8. The edit itself is usually correct but the arguments do not match the schema as the model invents made-up keys and Pi thus rejects the tool call and asks to try again.
That alone is not too surprising as models emit malformed tool calls sometimes. Particularly small ones. What surprised me is that this is getting worse with newer Anthropic models as both Opus 4.8 and Sonnet 5 show it but none of the older models. In other words, the SOTA models of the family are worse at this specific tool schema than their older siblings. — Read More
The brain’s language network is more extensive than previously thought
For decades, neuroscientists have known that specific regions in the brain’s left hemisphere are responsible for processing language. However, a new study by MIT researchers shows that language processing also occurs in many other parts of the brain.
Using functional magnetic resonance imaging (fMRI) data from more than 700 people, the researchers identified 17 additional regions of the brain that appear to play a role in language. These regions are scattered across the brain, including parts of the cerebellum, hippocampus, and cerebral cortex, and they make up about 5 percent of the total volume of the adult brain — about the size of a large strawberry. — Read More
Read the Paper
Context Window Management for Long-Running Agents: Strategies and Tradeoffs
Long-running agents are those capable of exhibiting sustained autonomous execution over time. In these agent-based applications — fueled by interactions with users or other systems in which information snowballs rapidly — the context window is a critical bottleneck. Agents and large language models, or LLMs in their abbreviated form, are two sides of the same coin in modern AI systems, so to speak. Accordingly, shifting from “LLMs as prompt-response engines” to “(agent-endowed) LLMs as long-running background processes” turns context windows into a major AI engineering bottleneck.
For all these reasons, managing context windows in the long run requires specific strategies like sliding windows, tiered memory, and dynamic summarization. This article presents five different operational strategies for this, together with their inevitable tradeoffs. — Read More
Rick's Cafe AI 