Sabitlenmiş Tweet
Mithril
106 posts
Mithril
@mithrilcompute
The AI omnicloud
Palo Alto & SF, CA Katılım Nisan 2023
18 Takip Edilen1.3K Takipçiler
Huge congratulations to our partners at Standard Intelligence! Thank you for using us as a compute partner.
Standard Intelligence@si_pbc
Computer use models shouldn't learn from screenshots. We built a new foundation model that learns from video like humans do. FDM-1 can construct a gear in Blender, find software bugs, and even drive a real car through San Francisco using arrow keys.
English
Mithril retweetledi
Computer use models shouldn't learn from screenshots.
We built a new foundation model that learns from video like humans do. FDM-1 can construct a gear in Blender, find software bugs, and even drive a real car through San Francisco using arrow keys.
GIF
English
Mithril retweetledi
We've opened new pools for on-demand, SPOT, and self-serve reserve (arbitrary durations, from hours to weeks) NVIDIA B200 GPUs on Mithril.
In general, these chips are hard to get access to, so we hope this helps!
Spot floor at $0.01 for long-running and flexible jobs.
Blackwells are really nice to work with. Having all the extra HBM is super convenient.
English
Researchers at the Broad Institute use Mithril’s GPU omnicloud to better understand gene expression.
Read how the Broad Institute is leveraging Mithril to accelerate biological discovery in our latest case study.
mithril.ai/blog/broad-cas…
English
Mithril retweetledi
Excited to see Anthropic's fast mode experiment. More options is always good — and counterintuitively, even 6x cost for 2.5x speed can make applications cheaper.
We ran an experiment at @mithrilcompute that shows why flexible compute economics are so powerful 👇
English
Mithril retweetledi
Introducing Locus: the first AI system to outperform human experts at AI R&D
Locus conducts research autonomously over multiple days and achieves superhuman results on RE-Bench given the same resources as humans, as well as SOTA performance on GPU kernel & ML engineering tasks.
RE-Bench is a collection of several frontier AI research tasks that typically take human experts (e.g., top ML PhDs and frontier lab researchers) several days. By scaling experimentation to far longer time horizons than previous systems, Locus represents a step change in AI scientist capabilities. 🧵
GIF
English
ARC Virtual Cell Challenge made possible by using Mithril
gmdbioinformatics.substack.com/p/arc-virtual-…
#mithril #mithrilcompute #mithrilai #arcinstitute #ml #machinelearning
English
Mithril retweetledi
A lot of the compute powering our AI models has been through @mithrilcompute
English
The only viable future for AI is non-autoregressive. It's upon this founding principle that we built Logical Intelligence — a company dedicated to making mission-critical software secure through AI formal verification.
We’re launching two AI agents + a new foundation model to deliver provably correct code, faster than ever before.
➡️ Non-autoregressive
➡️ Energy-based
➡️ Built for 100% mathematically precise reasoning
Unlike LLMs, our model doesn’t stumble piece by piece — it solves holistically, like magnets snapping a puzzle into place.
English
@logic_int Congrats @logic_int on the launch of Aleph and Noa! We're thrilled to be a trusted compute partner for you and the team.
English
We've had a ton of announcements this week at @mithrilcompute, but the real story is our CEO @jaredq_ out here making puns irresponsibly.
He must be stopped.
Jared Quincy Davis@jaredq_
We nearly named it MBTI (Mithril Batch Tokens Inference) last minute, so we could make Myers Briggs puns (although maybe that's more of a LinkedIn joke). Wrt "intelligence too cheap to meter", INTJ: I Need Trust in Jevon's*. If you're a pun enjoyer, comment or retweet with your MBTI/MyersBriggs and a pun, and we'll give 5 random people free tokens (500M/day) for 5 days! Some good ones we've heard—I Need Flops Pronto, I Need Tokens Please ---- *Jevon's paradox / the idea that if you make something 2x cheaper, demand will rise more than 2x to increase aggregate spend.
English
We nearly named it MBTI (Mithril Batch Tokens Inference) last minute, so we could make Myers Briggs puns (although maybe that's more of a LinkedIn joke).
Wrt "intelligence too cheap to meter", INTJ: I Need Trust in Jevon's*.
If you're a pun enjoyer, comment or retweet with your MBTI/MyersBriggs and a pun, and we'll give 5 random people free tokens (500M/day) for 5 days!
Some good ones we've heard—I Need Flops Pronto, I Need Tokens Please
----
*Jevon's paradox / the idea that if you make something 2x cheaper, demand will rise more than 2x to increase aggregate spend.
Jared Quincy Davis@jaredq_
Introducing Mithril Batch Inference A plug-and-play API built for massive-scale inference workloads! MBI leverages Mithril's global and dynamic omnicloud pool to push the frontier on: 🚀 Throughput: process 500 M tokens/day (higher on request) across models from OpenAI, Llama, Deepseek, Qwen, or custom models. 💰 Savings: Sufficiently low per M tokens we’re considering integrating @stripe's bridge network for fractions-of-a-penny "micro‑transactions"—we’re encroaching on the goal of "intelligence too cheap to meter". MBI is perfect for eval pipelines, embeddings & labeling, synthetic data generation, and more.
English
Read the full blog and get started → mithril.ai/blog/introduci…
English
Mithril retweetledi
Introducing Mithril Batch Inference
A plug-and-play API built for massive-scale inference workloads!
MBI leverages Mithril's global and dynamic omnicloud pool to push the frontier on:
🚀 Throughput: process 500 M tokens/day (higher on request) across models from OpenAI, Llama, Deepseek, Qwen, or custom models.
💰 Savings: Sufficiently low per M tokens we’re considering integrating @stripe's bridge network for fractions-of-a-penny "micro‑transactions"—we’re encroaching on the goal of "intelligence too cheap to meter".
MBI is perfect for eval pipelines, embeddings & labeling, synthetic data generation, and more.
English