San Francisco, CA
- Tweets 852
- Following 4,473
- Followers 612
- Likes 2,891
alignedai retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
863
2,605
19,653
4,374,039
alignedai retweeted
introducing muse wonder
our new frontier model. state-of-the-art at personal agents and professional agents.
our largest and most capable model yet—rivals the big bois
521
686
10,092
2,750,576
alignedai retweeted
MUSE WONDER
this is the culmination of over a year of research and infrastructure work in msl
since we originally shipped muse spark in april, we have continued to scale our models across pre-training and RL, now reaching the frontier 🏆
see our model's improvements on the artificial analysis index:
216
312
4,542
434,153
alignedai retweeted
Introducing Muse Wonder, our new family of frontier models with state of the art performance for personal and professional agentic work.
2,371
3,077
38,845
5,965,933
Worked on project hatch (muse) - love the reception @Muse is receiving. People are slowly realizing it is model + harness + UI/UX that maximizes utility to the end user.
1
23
1,365
alignedai retweeted
the most underrated part of muse is the model that powers it. thank you to the MSL research team—some of the most technically brilliant, hardworking scientists and engineers i’ve ever known. you brought original ideas to difficult problems, invented new methods, and battled through enormous technical complexity to build the jewel at the heart of muse.
thank you to the teams working on agentic post-training, tool and computer use, memory, personalization, alignment, safety, and multimodal understanding. thank you to the people building data, evals, and infrastructure beneath it all. so much scientific ingenuity lives in those details: designing a better experiment, finding the flaw in an apparently good result, staying with a stubborn problem until a new way forward becomes clear.
and now that jewel powers a beautiful product your grandma can use without knowing a thing about the extraordinary complexity underneath. it will just work for her. all those discoveries, all that patient work, becoming something simple and useful in another person’s hands. there is something magical about that. i’m grateful to be building with you.
74
98
1,559
205,886
alignedai retweeted
Muse for Mac is out today! It works across apps, files, calendar, notes, and messages on your computer. You control what it can access. The team is shipping fast. Download at ai.meta.com/muse/download
951
577
9,393
961,256
#1 on AA-WER Streaming.
Accuracy: 3.1% WER 14% lower than ElevenLabs Scribe v2 Realtime.
Latency: 160ms finalization 2.7× faster than Cartesia Ink-2 semantic endpointing.
Price: $0.18/hour 2.2× cheaper than ElevenLabs/Deepgram.
Streaming: 80ms audio chunks.
Languages: 70+ trained, 25 extensively verified.
A very strong accuracy × latency × cost frontier.
Meta has released Muse Voice Transcribe, taking the #1 spot for Final Transcript accuracy on AA-WER Streaming with 3.1% WER at 0.16s after end of speech
Muse Voice Transcribe is the first streaming Speech to Text model developed by Meta Superintelligence Labs. Meta states that the model was trained on more than 70 languages, with 25 extensively verified, and supports audio inputs exceeding one hour without required post-processing. It processes audio in 80ms chunks and is available through the Meta Model API, Meta AI for Mac and Muse Code.
Key takeaways
➤ Final Transcript: Muse Voice Transcribe achieves 3.1% WER at 0.16s after end of speech. It is more accurate and faster than Cartesia Ink-2 (semantic endpoints) at 3.4% and 0.43s, and more accurate but slower than Cartesia Ink-2 (external endpoints) at 4.0% and 0.07s. It is also more accurate, though slightly slower, than ElevenLabs Scribe v2 Realtime at 3.6% and 0.14s.
➤ First Partial Transcript: The model achieves 3.6% WER at 0.13s, just ahead of ElevenLabs Scribe v2 Realtime on accuracy and latency. It is more accurate and faster than Cartesia Ink-2 (semantic endpoints) at 4.9% and 0.17s, and more accurate but slower than Cartesia Ink-2 (external endpoints) at 4.0% and 0.07s.
➤ Price: Muse Voice Transcribe costs $0.18 per hour, or $3 per 1,000 minutes. This is below Cartesia Ink-2 at $4 and less than half the $6.50 charged for ElevenLabs Scribe v2 Realtime and Deepgram Flux.
See more details below ⬇️
1
23
1,860
This is incredible @AIatMeta strikes again
Muse Voice Transcribe is MSL's first real-time audio perception model -- rolling out today. SOTA in streaming speech-to-text, it handles speaker diarization, and endpointing natively in a single model.
2
24
1,264
alignedai retweeted
1/ today we're rolling out muse voice transcribe, our first real-time audio perception model - SOTA in streaming speech-to-text. also handles speaker diarization and endpointing natively in a single model.
83
149
2,000
163,722
Mac app link: ai.meta.com/meta-ai/assistan…
We launched the Meta AI Mac OS app today! 🚀
I particularly love the dictation feature which allows me to dictate anywhere on my computer with ultra high accuracy. Just hold down 'fn' and yap!
7
631
alignedai retweeted
Introducing Muse Glimmer, an open-weight 30B-parameter model optimized for local, always-on agent workflows.
Muse Glimmer delivers strong performance on key agentic use cases and benchmarks compared with leading models in its size category, and is designed to run entirely on consumer hardware like a Mac or PCs with performant GPUs.
In keeping with our long tradition of sharing fundamental AI research, we’re releasing model weights under a permissive Apache 2.0 license.
🧵👇
377
1,099
8,237
1,437,202
Today we released Muse Code, a terminal coding agent, powered by our new Muse Spark 1.2 model. Learn more 👇
Releasing Muse Code in beta today. It's a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results. Powered by Muse Spark 1.2, a coding-focused model update.
581
111
879
284,779
Muse Spark 1.2 by @AIatMeta is in the Agent Arena!
Bring your toughest prompts to power the leaderboards.
In Agent Arena, we measure models on millions of real-world, long-horizon agentic tasks from a global community of users. Models can access web search, filesystem, and terminal tools to complete complex workflows. The leaderboard measures model performance on outcomes relative to the average model using a causal tracing methodology.
In addition to Agent Arena, Muse Spark 1.2 is available in Text, Vision and Code Arena: Frontend.
Introducing Muse Code (beta), a terminal coding agent built for long-horizon software engineering, powered by our new Muse Spark 1.2 model.
Muse Code plans, implements, and validates complex, multi-file changes across large repositories with persistent sub-agents that solve difficult problems faster, more accurately, and with less intervention.
🧵👇
9
13
240
25,835
alignedai retweeted
Muse Spark 1.2 is available today in Muse Code and in Meta Model API. App coming soon
1
1
2
553
Tie for 3rd among US labs for now 👀🫡👀
Meta has released Muse Spark 1.2. It's their third release in four months and scores 54 on the Artificial Analysis Intelligence Index, significantly improving agentic knowledge work capabilities over prior releases and putting Meta next to SpaceXAI in a tie for third place amongst US labs
Muse Spark 1.2 (xhigh) lands at 54, up 3 points from Muse Spark 1.1 (51) and 11 points from Muse Spark 1.0 (43, April). It enters effectively tied with GPT-5.5 (xhigh, 55) and Grok 4.5 (high, 54), narrowly behind current frontier models Claude Opus 5 (max, 61), Claude Fable 5 (max w/ fallback, 60), GPT-5.6 Sol (max, 59), and Kimi K3 (max, 57)
Congratulations to @AIatMeta, @finkd, and @alexandr_wang on the release!
Key Takeaways:
➤ Muse Spark 1.2 gets closer to the frontier on agentic knowledge work. At Muse Spark 1.1's launch, we noted agentic knowledge work as its clearest gap; Muse Spark 1.2's gains help to close this. Its GDPval-AA v2 Elo rose 260 points to 1631, #5 among all models we have benchmarked and ahead of Claude Opus 4.8 (max, 1588). Terminal-Bench 2.1 gained 2 points (78% to 80%), and Tau3-Bench Banking rose 2 points (25% to 27%)
➤ Among the most cost-efficient models at its intelligence level. Muse Spark 1.2 costs $0.40 per Intelligence Index task at Meta's unchanged $1.25/$4.25 per 1M token pricing, with only Grok 4.5 (high, $0.37) and GPT-5.6 Sol (medium, $0.39) cheaper in its intelligence cluster - GPT-5.6 Terra (max, $0.51), Kimi K3 (max, $0.86), and GPT-5.5 (xhigh, $1.18) all cost more per task. The cost increase over Muse Spark 1.1 ($0.29 per task) is driven by increased token usage per Intelligence Index task
➤ AA-Omniscience abstention rate increases. The score rose from 18 to 22 as the hallucination rate fell 10 points (38% to 28%) and the attempt rate dropped from 82% to 67%. This heavy abstention (not answering questions when unsure) now drives both the low hallucination rate and a lower accuracy (41% to 38%)
➤ Scientific Reasoning results remain largely unchanged. CritPt notably gained 3 points (15% to 18%), while SciCode fell 2 points (58% to 56%), and Humanity's Last Exam fell 1 point (45% to 44%)
Other model details:
➤ Context window: 1M tokens, unchanged from Muse Spark 1.1
➤ Pricing: unchanged from Muse Spark 1.1: $1.25/$4.25 per 1M input/output tokens, with cache hits discounted to $0.15 per 1M
➤ Availability: Meta's first-party API at launch
3
38
3,449
alignedai retweeted
Meta has released Muse Spark 1.2. It's their third release in four months and scores 54 on the Artificial Analysis Intelligence Index, significantly improving agentic knowledge work capabilities over prior releases and putting Meta next to SpaceXAI in a tie for third place amongst US labs
Muse Spark 1.2 (xhigh) lands at 54, up 3 points from Muse Spark 1.1 (51) and 11 points from Muse Spark 1.0 (43, April). It enters effectively tied with GPT-5.5 (xhigh, 55) and Grok 4.5 (high, 54), narrowly behind current frontier models Claude Opus 5 (max, 61), Claude Fable 5 (max w/ fallback, 60), GPT-5.6 Sol (max, 59), and Kimi K3 (max, 57)
Congratulations to @AIatMeta, @finkd, and @alexandr_wang on the release!
Key Takeaways:
➤ Muse Spark 1.2 gets closer to the frontier on agentic knowledge work. At Muse Spark 1.1's launch, we noted agentic knowledge work as its clearest gap; Muse Spark 1.2's gains help to close this. Its GDPval-AA v2 Elo rose 260 points to 1631, #5 among all models we have benchmarked and ahead of Claude Opus 4.8 (max, 1588). Terminal-Bench 2.1 gained 2 points (78% to 80%), and Tau3-Bench Banking rose 2 points (25% to 27%)
➤ Among the most cost-efficient models at its intelligence level. Muse Spark 1.2 costs $0.40 per Intelligence Index task at Meta's unchanged $1.25/$4.25 per 1M token pricing, with only Grok 4.5 (high, $0.37) and GPT-5.6 Sol (medium, $0.39) cheaper in its intelligence cluster - GPT-5.6 Terra (max, $0.51), Kimi K3 (max, $0.86), and GPT-5.5 (xhigh, $1.18) all cost more per task. The cost increase over Muse Spark 1.1 ($0.29 per task) is driven by increased token usage per Intelligence Index task
➤ AA-Omniscience abstention rate increases. The score rose from 18 to 22 as the hallucination rate fell 10 points (38% to 28%) and the attempt rate dropped from 82% to 67%. This heavy abstention (not answering questions when unsure) now drives both the low hallucination rate and a lower accuracy (41% to 38%)
➤ Scientific Reasoning results remain largely unchanged. CritPt notably gained 3 points (15% to 18%), while SciCode fell 2 points (58% to 56%), and Humanity's Last Exam fell 1 point (45% to 44%)
Other model details:
➤ Context window: 1M tokens, unchanged from Muse Spark 1.1
➤ Pricing: unchanged from Muse Spark 1.1: $1.25/$4.25 per 1M input/output tokens, with cache hits discounted to $0.15 per 1M
➤ Availability: Meta's first-party API at launch
62
117
1,159
305,753