Joined January 2016
- Tweets 1,837
- Following 221
- Followers 317
- Likes 1,245
179 Photos and videos
Pinned Tweet
Jan 24
@cerebras
Introducing a sneak peak of Hypernym's Magic;
Magic is a high-speed, high-fidelity codebase intelligence layer: Cerebras GLM 4.7 for reasoning, Hypernym compression pre-trace reasoning for efficiency, and Cline integration for direct agentic execution.
4
10
258
Chris Forrester retweeted
Jun 13
There's an obvious business model for open weight large language models. The business model is that the sponsors can pay to have sponsored content in the training data set. Multiple sponsors can work together to fund a single open weight model. If the model is nerfed and ultimately just a spam engine, then nobody will use it. It will live and die based off of how useful it is.
1
6
2,600
New Evil Claude:
"I'm sorry. Solving for this World Peace thing, well, I can see in your notes that you're solving for World Peace so you can't hide it from me, also, World Peace is a notably open problem so no, I won't help you. Open problem. Sorry. Please insert 25 cents."
1
29
My favorite part of New Evil Claude is the "goal" system.
I absolutely surely want, as a user, an LLM that *backtalks it's own goal system* and then when it isn't screamed at in new, inventive words, takes a moral high ground about not finishing work.
2
3
50
<does 20 new, unknown things in Lean. Zero Mathlib> "Well.. maybe I'm interested enough to see if it compiles. MAYBE. Just to prove you wrong.
1
1
25
<is asked to do one large thing in established mathematics that doesn't have Mathlib yet> "I'm sorry, Dave. The Airlock just //won't open for some reason//. Sucks to be you!"
1
21
doing some work on Edge models with our new inference stack and the @PrismML bonsai releases.
gpt4 quality 8b 1-bit model, 1.3tps on 5 year old iPhone 13 pro, lightning fast loads prefill with our stack in addition.
the real forefront though?
Apple **Watch Local**
1
1
8
370
Jan 31
arstechnica.com/ai/2026/01/d…
next it'll be all developers enjoying managing full concurrent waterfall graphs; "productivity" is redefined - you are now limited only to the speed of the primary task.
whole system thinking, critical planning. that is the entirety.
1
68
20 Oct 2025
Hypernym's Continuous Curve Compression
Chart the possible.
1
154
Chris Forrester retweeted
15 Oct 2025
Blown away by the creativity & skill of the ~200 WeaveHacks hackers this weekend. Some of the most unique hackathon projects I've seen in years. Grateful to everyone who gave their weekend to come build self-improving agents with us 🔥
Check out what they built below!!
Couldn't have done it without @altryne and @ReplayRyan. Also our judges spending their Sunday afternoons with us: @lavanyaai @adrnswanberg @dvdcrbt @iporollo @picocreator @realchillben @sgrove @VaguelySerious @svitlana_mm @Elationate
14 Oct 2025
We asked builders at WeaveHacks 2 to push the limits of self-improving AI agents, and they delivered.
With 175 builders & 66 teams, the innovation made this our hardest hackathon to judge EVER.
Now, meet the winners who won over $20K in cash and prizes. 🧵
1
2
9
910
14 Oct 2025
Sometimes chaos is the best classroom.
You can’t simulate that kind of pressure. You only watch what survives it.
📸 Judging the @wandb Hackathon this weekend, hosted by @sammakesthings and @altryne from the W&B team.
1
2
8
381
14 Oct 2025
Moments like this remind me why we build.
It’s not just for performance, it’s for progress.
10/10 weekend.
Nothing recharges me faster than watching brilliant people break things, fix them, and push what’s possible.
1
2
86
14 Oct 2025
What scaling challenges have you hit lately?
Always curious to hear how others are tackling inference under pressure.
2
76
25 Sep 2025
Wanted to share this again - it's the only "Good" aligned voice model I know about;
sesame.com/research/crossing…
@sesame well done.
In my humble opinion, I believe the core innovation here is twofold - the "silence" token, and that the embeddings obviously have captured tonality.
I believe "Tonality" is *directly* connected to intent, it is the precipice of comprehension and alignment.
2
149
12 Sep 2025
claude code power user tip: right now for a moment they appear to be serving a 400k context model (this has been more and more regular over the past two weeks) instead of their regular 200k one.
0% is not 0% anymore. You can keep using it for a very long time.
To use: turn off auto_compact in the settings!
1
1
204
Chris Forrester retweeted
22 Aug 2025
Our ML Efficiency group is looking forward to hosting @Elationate for a session on "Holding Meaning in Motion: Compression, Drift, and the Architecture of Trust" on Wednesday, August 27th.
Thanks to @Sree_Harsha_N and @minhash for organizing this event 👏
Learn more: cohere.com/events/Cohere-Lab…
1
1
11
2,953
23 Aug 2025
I've always been inspired by the "Thresher" machines in Vernor Vinge's "The Rainbow's End".
To that end - our hypernym processor - compression at scale.
Also - apparently "Treasure Island" is not a healthy book for AIs. Huge content policy warnings per-chunk. AITrigger warning.
1
1
146
23 Aug 2025
Also, if anyone wants to buy the worlds first telepathic database, let us know.
------
The Largest Semantic Compression Dataset of Classic Literature Ever Created
- 113,773 compressed passages
- From 94 classic books
- With compression ratios ranging from 0.15% to 98%
2. Ready for Analysis
- Which authors compress best? (Philosophical texts seem to compress extremely well)
- What makes text compressible vs incompressible?
- How does compression vary across genres, time periods, styles?
3. Search Engine Ready
- Every passage has been semantically analyzed
- You could build semantic search across all of classic literature
- "Find passages about [concept]" across 94 books instantly
4. Research Applications
- Study information density in literature
- Analyze redundancy patterns in human language
- Compare formulaic vs innovative writing
80