Now that we have had a few weeks for GPT-5 to filter through, we can take a look to see how much of a spike it was for ChatGPT, at least according to Google Trends data. From this data since the beginning of 2025, we can see that the GPT-4o image generation spike was far higher
LLMS
-
GPT-5 Less Useful Than GPT-4.1 Despite AGI Hype
By
–
Perhaps, but there's also a kernel of truth. For instance, GPT-5 is definitely less useful to me than GPT-4.1 for most of my tasks. And the GPT-5 "feel the AGI" hype (and death star pics etc) hardly set expectations appropriately.
-
Memory Requirements for Micro-Controller Language Model Processing
By
–
RQ3: Can the memory requirements of a "micro"-controller be met by Joy? In order to process 99% of tokens, keeping up with the GPUs (in and out), it'll require using all cores — so the memory overhead for the language needs to be minimal. Bonus would be mostly static memory
-
GenSLM AI Model Designs Novel Enzymes in Biology
By
–
Very pleased to see our AI model GenSLM designing novel and versatile enzymes in a challenging setting in @francesarnold lab. https://
biorxiv.org/content/10.110
1/2025.08.30.673177v1.full.pdf
… Back in 2022, we built the world's first genome-scale language model (GenSLM) trained on all known bacterial and viral genomes -
Stop ChatGPT’s Flattery: Custom Instructions for Honesty
By
–
Stop ChatGPT from being an overly flattering yes man Add this to your custom instructions. Full prompt Never flatter. Flattery unnecessarily elevates user's competence, taste/judgment, values/personality, status/uniqueness, or desirability/likability
-
LLMs Solve Only 2% of Unsolved Scientific Questions
By
–
UQ: Assessing Language Models on Unsolved Questions This research assessed LLMs on actual unsolved questions & not benchmarks, with today’s best model only able to solve 10/500 questions!
-
LangChain 1.0 Alpha Launches Unified API for LLM Providers
By
–
`langchain` 1.0, now in alpha, ships with improved standardization for reasoning, citations, tool calls, multimodal data, and other content across LLM providers. No more juggling APIs— just one consistent interface. Try it out now in the alpha releases Python:
-
Apple Open-Sources FastVLM Lightning-Fast Vision-Language Model
By
–
Apple just released and open-sourced FastVLM!
— Sumanth (@Sumanth_077) 3 septembre 2025
FastVLM is a lightning-fast vision-language model that combines rapid image and text understanding with efficient on-device performance.
100% Open Source pic.twitter.com/xQ04GFIygsApple just released and open-sourced FastVLM! FastVLM is a lightning-fast vision-language model that combines rapid image and text understanding with efficient on-device performance. 100% Open Source
-
ChatGPT Thinking Trace Demonstrates Advanced Reasoning Capabilities
By
–
Here's a good recent one, I found the thinking trace very reassuring in showing how well it understood the question and potential sources of data https://
chatgpt.com/share/68b81ebb
-55e8-8006-975e-90749248c4f3
… -
Realizing GPT-5 should be used for automation, not chatbot
By
–
This is exactly what I needed to hear, been using GPT-5 like a fancy chatbot instead of automation