What if your LLM could remember everything without rewriting its entire memory? Researchers from NTU, Fudan, and Mind Lab present δ-mem. It’s a tiny memory module that compresses past info into a fixed-size matrix, updated via a simple delta rule. This matrix then tweaks the
LLMS
-
GLM 5.2 weights backed up across several nodes
By
–
GLM 5.2 weights are downloaded and backed up across several nodes They can never take away my Fable 5
-
Speculative Decoding progress since 2023 brings joy
By
–
The state of Speculative Decoding nowadays makes me extremely happy
— Ahmad (@TheAhmadOsman) 20 juin 2026
We came a long way from 2023 / 2024 in this area https://t.co/HPoc9xMY3d pic.twitter.com/Lw1gYvzz1cThe state of Speculative Decoding nowadays makes me extremely happy We came a long way from 2023 / 2024 in this area
-
Codex and Fable 5 interact via plugin with Claude Code
By
–
— Aaron Ng (@localghost) 19 juin 2026
Here is Codex talking to Fable 5 with my plugin. Unlike the API call, `claude-channel-cli` talks to the same full harness of Claude Code as you. This means you can talk to either AI & they will guide each other naturally with
-
Google’s AI turnaround short-lived as OpenAI and Anthropic reclaim lead
By
–
This is such a major falling off. After being mocked for years for being a total AI laggard, Google had a major turnaround at the end of the last year with Gemini 3 and Nano Banana 2 releases. But OpenAI and Anthropic quickly caught up, took over the lead, and then gradually
-
Gary Marcus argues LLM job loss hype is overblown
By
–
Only a narrative violation if you believed people with a lot of money at stake. • Most of the people that said lots of jobs would soon disappear were running LLM companies trying to overhype their wares. • Personally, I told you that the hype wasn’t likely true (e.g. in my
-
Claude message replaces SQL for trivial text edits
By
–
Probably even better now that trivial edits to it are a message to Claude away versus having to write a SQL query to do equivalent of a one character/line insert/edit.
-
China unbeatable in token prices against the West
By
–
China is simply unbeatable in terms of token prices and plays in a completely different league than Western companies and models.
-
Exponential Progression and Clustering of AI Actors
By
–
Another exponential, that of Artificial Analysis. It is that almost the entire field is progressing upward at the same time. OpenAI, Anthropic, Google, xAI, Meta, DeepSeek, Alibaba, Mistral, Kimi and others are now *much more clustered* than they were a…
-
dcode: model-agnostic harness, recommended with Fireworks
By
–
It is indeed very good! Do not try it in claude code/codex – these harnesses are too tailored for their proprietary models. dcode (deepagents code) is a model-agnostic harness – try it there with @FireworksAI_HQ: dcode –model