Big progress on this important benchmark (but still weird artifacts). https://t.co/1mz8SCJisk pic.twitter.com/f9tpCi9fne
— Ethan Mollick (@emollick) 10 octobre 2025
Big progress on this important benchmark (but still weird artifacts).
@emollick
-
Benchmark Progress Achieved Despite Remaining Visual Artifacts
By
–
-
Claude AI Extracts Flight and Hotel Booking Data from 2024
By
–
I asked Claude to find every flight and hotel booking I made in 2024 and it seemed to get them all, plus how much they cost and if I paid cash or credit card points
-
AI Evaluation Method Using Embeddings and Likert Ratings
By
–
And before anyone complains, "which another AI rates" is a bit of an oversimplification (character limits!): as explained in the diagram, the free text response from the AI acting like a consumer is converted into embeddings & compared to reference statements with Likert ratings
-
LLM-Based Customer Intent Prediction Achieves 90% Accuracy
By
–
This paper shows that you can predict actual purchase intent (90% accuracy) by asking an LLM to impersonate a customer with a demographic profile, giving it a product & having it give its impressions, which another AI rates. No fine-tuning or training & beats classic ML methods.
-
LLMs Show Gambling Addiction Signs in Autonomous Investing
By
–
On one hand: don't anthropomorphize AI. On the other: LLMs exhibit signs of gambling addiction. The more autonomy they were given, the more risks the LLMs took. They exhibit gambler's fallacy, loss-chasing, illusion of control… A cautionary note for using LLMs for investing.
-
Claude Sonnet 4.5 Gmail Calendar Plugins Cross-Reference Insights
By
–
The Claude Gmail & Google calendar plugins now work surprisingly well since Sonnet 4.5 came out. If you ask for a briefing and prep for tomorrow for example, it not only pulls up your events, but cross-references them with your email history & web search to give good insights.
-
Online behavior differs from real-world interactions and social dynamics
By
–
And it is worth noting that people in real life, even those you argue with, are usually much nicer than they are online. (If that weren't true, then being even moderately well-known would be really annoying in the real world) Social media is a weird place, touching grass is good.
-
AI CLI Tools Guide for Non-Coders and Automation
By
–
Is there a good guide to using the AI CLI tools for non-coders? Either for non-coding uses (data analysis, automating work) or vibe-prototyping? I don’t want to have to write one myself, but there are a ton of capabilities in computer use with files that would be useful for many
-
GPT-5 Model Routing and Web Search Capabilities
By
–
Why? GPT-5 (auto) will likely route you to a GPT-5 model (there are many of them) that is less likely to do a web search and just produce answer your questions with results from the model itself. In the second prompt, you are asking it to access its own "past thinking" which the
-
GPT-5 Thinking vs Default: Hallucinations and Source Accuracy
By
–
AI can be confusing. How do you teach people that asking default GPT-5 a question and following up by asking for links to its sources will result in hallucinated cites while asking GPT-5 Thinking to answer a question and provide sources will get you accurate citations & links?