By
–
Unappreciated fact is the second scaling law does not seem to completely plateau in many tasks: throw more tokens at a reasoning AI model and get better answers, especially with a simple harness. Benchmark performance is actually limited by token usage. https://
open.substack.com/pub/joelbkr/p/
many-benchmarks-scores-would-appear?r=i5f7&utm_medium=ios
…
By
–
Gemma 4 E4B is impressive for an on-device LLM. GPT-4ish quality, and expect hallucinations.
— Ethan Mollick (@emollick) 5 avril 2026
Here is: “List five sociological theories starting with u and what they are. Then describe them in a rhyming verse”
Its in real time, the last is a little bit of a stretch, but not bad! pic.twitter.com/nJ5HuQsFSB
Gemma 4 E4B is impressive for an on-device LLM. GPT-4ish quality, and expect hallucinations. Here is: “List five sociological theories starting with u and what they are. Then describe them in a rhyming verse” Its in real time, the last is a little bit of a stretch, but not bad!
By
–
Big deal paper here: field experiment on 515 startups, half shown case studies of how startups are successfully using AI. Those firms used AI 44% more, had 1.9x higher revenue, needed 39% less capital:
1) AI accelerates businesses
2) The challenge is understanding how to use it
By
–
Need to set up my OpenClaw to update and restart my Claude Dispatch to add computer use so I can use that instead.
By
–
This new Nature paper (using old models) illustrates the point of my latest Substack post on AI interfaces. AI did a good job diagnosing medical issues, but when users had to interact with chatbots the interface led to confusion & worse answers My post: https://
oneusefulthing.org/p/claude-dispa
tch-and-the-power-of
…
By
–
From the replies it is clear there is still no agreement what RAG is, so I guess the truth of my statement depends on your definition of RAG.
By
–
Here’s an independent domain extension of METR’s famous time-horizon analysis, applying it to offensive cybersecurity with real human expert timing data Similar to METR: 5.7 months doubling time. Frontier models now succeed 50% of the time at tasks that take human experts 10.5h.
By
–
The RAG era was short-lived, but intense. (Not that RAG is not useful, but it is no longer the dominant paradigm for supplying context to agents)
By
–
Gemini was the only frontier model that was susceptible to these sorts of prompt injections (Though we tested Gemini 3 & not 3.1)
By
–
New report from us: Can you prompt inject your way to an “A”? As LLMs increasingly are used as judges, people are inserting AI prompts into letters, CVs & papers. We tested whether it works. It does on older & smaller models, but not on most frontier AI: https://
gail.wharton.upenn.edu/research-and-i
nsights/hidden-prompt-injections/
…