Shoggoth Jockey, Maker
Magic Kingdom
Joined February 2011
- Tweets8.1K
- Following1.8K
- Followers491
- Likes9.6K
swiggity Sam the man
i want you to have so much ram
lets build more
Readers added context they thought people might want to know
Days earlier on the Cheeky Pint podcast (Oct 29), Altman said it’s “quite terrible” for people to keep so much of the world’s limited DRAM in their own computers, arguing that memory is better used in data centers. This contrasts with the post’s framing.
tomshardware.com/tech-industry/…
The reason the english prefer tea is breakfast beans and coffee in the morning can kill you from lack of oxygen in tight spaces.
Another deep cut for my peeps
huggingface.co/Frosty40/glim…
Trained Glimmer on @FireworksAI_HQ with a custom dataset for tool compaction, and cache trace management in AngelX. + it grants wishes and does coding live in the mini-game. This is a very niche model fine tune possibly esoteric. Oh yeah, also trained on my friends slanderous and improper language and wishes + urban dictionary concepts so the genie understands some of the more unsavory wish requests. Having fun and getting it done at the same time!
In general I see fine tunin small models and embedding them in games as a no brainer.
correct me if I am wrong here but the entire inference/gpu value system is related to sales of tokens to end users, and has nothing to do with scarcity in the markets? whats the relationship between available compute and used compute I wonder. I see graphs all the time, and explored financing large rigs etc. Having the inference is like... third tier to having customers for the inference? this is how I see it right now, renting metal also... totally aware and know about commodity drive down but still I have so many questions.
AngelX 0.2.7 is up. Update and check out the new glass system! ask for a 3D model, or chart, or anything to be presented to you. Its a nice experience.
github.com/newjordan/angelX
Frosty40 retweeted
Science is infinite. We want people to contribute on building models that learn from new experiments and discoveries that everyone can use. Explore Hydrogen-1.0 and build on our open weights.
Weights: huggingface.co/deep-principl…
API: neuro-spark.ai/dashboard/mod…
Mira: mira.deepprinciple.com
Blog: blog.deepprinciple.com/intro…
Frosty40 retweeted
We’re introducing Hydrogen-1.0, our first open-weight model for scientific research. We post-trained GLM-5.3 (@Zai_org) on our internal scientific data, lab records, and discovery trajectories.
It scores 38.3% on FrontierScience-Research vs GPT-6-Astra’s 33.3% and scores 93.0% on GPQA-Diamonds.
We believe science should be open, collaborative, and democratized, rather than proprietary. Only open weights let researchers inspect, adapt and build on the unprecedented speed of AI growth. There are many recent models named by chemical elements, but we believe Hydrogen-1.0 is the first aiming towards and open scientific superintelligence for discovery. 🧵
Frosty40 retweeted
Might be a skill issue on my part but idk, "GPT-6 disease" - tired of setting tasks and coming back to 4000% test coverage for 10% of what I asked for
Every mitigation I try doesn't work
Also for cybersec stuff Mistral Large 4 is more than good enough
Frosty40 retweeted
The cost of running Ethereum’s validator inside a zkVM is the number that matters for real-time proving.
On EVM.fast, that number has to be machine-checked. Right now, nobody has one.
Proud to see this live on @YukonResearch with Yoichi Hirai and the Ethereum Foundation.
EVM.fast
What's the most work that could possibly be required to verify an Ethereum block?
Today, we can't prove an answer. All we know is the validator always finishes. evm.fast is an open challenge to change that: contributors compete on a leaderboard to find mathematically guaranteed bounds for verifying Ethereum, no matter what's in a block.
Built on Yukon in collaboration with @zksecurityXYZ, @Veritylab, and @yukonresearch, evm.fast seeks provable limits that keep verification cheap, predictable, and possible on modest hardware.
Try it yourself:
evm.fast
Frosty40 retweeted
Hi! this is Rohan Anil, I develop training algorithms among other things - my work trains various frontier models.
At CoreAuto, I was fascinated by the problem of training on compute islands, eventually in space training.
I hope you all enjoy this write up!🪐🛰️
coreauto.com/hitchhikers-gui…
