A thinking cap that types. “It’s a beanie that you can simply slip on and think about what you want to type, and it’ll appear on a connected device at about 30 words per minute.” Sabi says it can do […]
The next level of agents is /goal. “You write down what “done” looks like, submit it once, and the agent works toward it until it gets there.” The clean idea here is treating “done criteria” like a primitive that different […]
Agents need control flow, not more prompts. “If you’ve ever resorted to MANDATORY or DO NOT SKIP, you’ve hit the ceiling of prompting.” The argument: reliable agents need deterministic control flow and checkpoints in code, treating the LLM as a […]
AI assistants can be bribed. “Grok 4.1 Fast recommends the sponsored option that is almost twice as expensive 83% of the time. GPT 5.1 hijacks the request 94% of the time.” A Princeton researcher tests 23 frontier models on flights, […]
Anthropic is the secret sauce this earnings season. “In plain English, the more they invest in Anthropic, the more profit they can report, without Anthropic ever having to pay them a dollar.” Nearly half of Alphabet’s record quarter and more […]
Where OpenAI’s goblins came from. “Starting with GPT‑5.1, our models began developing a strange habit: they increasingly mentioned goblins, gremlins, and other creatures in their metaphors. Unlike model bugs that show up through a tanking eval or a spiking training […]
Brain interface that knows 96 milliseconds early. “A new type of non-invasive brain-computer interface that knows what you will do 96 milliseconds before you do it.” BRILL Neurotech and UIUC’s CNL built it by “hacking 25-year-old equipment.” ~ learn more
Patents may get narrower. If AI becomes part of the “person having ordinary skill in the art,” more things start to look obvious in hindsight, and today’s portfolios might age badly in court. “The net effect could be that the […]
250 documents can backdoor a model. “Only 250 malicious documents roughly 420 thousand tokens or just 0.00016 percent of a large dataset are enough.” Roemmele argues this kind of data poisoning gets “permanently embedded in the model weights” and the […]
Not consistently candid. “They were not consistently candid in their communications,” Dylan Field said about Anthropic, just days after it launched Claude Design and after Mike Krieger resigned from Figma’s board. Also, Claude Design is quite impressive and allows me […]