Applied Scientist @ Amazon Alexa AI
United States
Joined June 2016
- Tweets74
- Following248
- Followers199
- Likes240
Devamanyu Hazarika retweeted
1.4.8 made Muse Code a lot more token-efficient, so we wanted you to feel it right away.
🎉 Everyone's usage is reset, effective now. Have fun.
Muse Code 1.4.8 is out and is focused on token efficiency.
- Auto-review and background agents now use a cheaper model when available, and long auto-reviews stop sooner instead of burning tokens.
- We've fixed a bug that routed a small number of users who had enabled contributor mode to the full-priced models.
Devamanyu Hazarika retweeted
Muse Spark Open -- weights releasing today -- runs great on new Mac Studios thanks to work with @johnternus and Apple.
Getting to work with this super talented team and shipping magical stuff for everyone. What else can one ask for!
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.
Devamanyu Hazarika 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.
Devamanyu Hazarika retweeted
Muse Code users:
1/ we had a recent issue with subscription usage metering that we debugged through the weekend. for every request, we were reserving a pool of tokens to complete the request. necessary retries weren't being processed correctly, leading to request failures as 429s and loss of unused tokens.
2/ contributor tier limits have now been raised significantly on every subscription tier to provide for more generous usage moving forward! would love to see how far you guys are able to get on a sub now.
we have now reset everyone's usage for the metering fix + the new elevated limits 🥳
Devamanyu Hazarika retweeted
Here is a walkthrough of Muse! Excited to share more about the personal agent we’ve been working on.
muse.ai
Our new Muse ai is live!
It has been my goto personal AI assistant and helps manage many aspects of my life. Can't wait for everyone to give it a try!
Try the magical experience at: muse.ai
research.meta.ai/blog/securi…
Working on improving coding capabilities of this model has been a surreal experience.
Try it out!
underrated result of muse spark 1.3
try it out with muse code:
curl -fsSL dev.meta.ai/install.sh | bash
Devamanyu Hazarika retweeted
1/ today we’re releasing muse spark 1.3—available in muse code & the meta model api.
this is our most capable model yet—frontier performance almost too cheap to meter. much stronger at agentic and coding with better usability. we think users will really notice the jump.
Devamanyu Hazarika retweeted
Muse Spark 1.3 is rolling out today with frontier performance almost too cheap to meter. This is the biggest jump we've made so far on coding and agentic work. Try it in Muse Code and our API.
Next up 🍉 and Muse Spark open weights releases coming soon.
Devamanyu Hazarika retweeted
Muse Code is out of beta and now built to handle bigger, more complex engineering tasks. Developers can get started with one command today:
curl -fsSL dev.meta.ai/install.sh | bash
Checkout our new Muse Spark 1.1 (available with an API)!
We have made huge leaps in its agentic coding capabilities.
Excited to see how the world builds on it!
Devamanyu Hazarika retweeted
Another release this week🔥
Muse Spark 1.1 is a considerable upgrade in capability, especially for agentic tasks. Plus… It’s available via API!
We see this model as a workhorse: fast and very reasonable priced, with strong all-round capability.
Devamanyu Hazarika retweeted
(1) Today we're releasing Muse Spark 1.1 -- a strong agentic and coding model at a very low price. It's available through our new Meta Model API and in Meta AI.
Devamanyu Hazarika retweeted
1/ muse spark 1.1 is an industry-competitive agentic and coding model. across many agentic evals it rivals gpt-5.5 and opus-4.8.
available now through the new meta model api and in meta ai. 🧵
Devamanyu Hazarika retweeted
1/ releasing muse image today — the first image generation model from MSL. it's agentic: pairs with muse spark to reason through your prompt, search the web, and plan before it generates. people get what they meant on the first try. live now in the Meta AI app.
Devamanyu Hazarika retweeted
Excited to release PrefEval (ICLR '25 Oral), a benchmark for evaluating LLMs’ ability to infer, memorize, and adhere to user preferences in long-context conversations!
⚠️We find that cutting-edge LLMs struggle to follow user preferences—even in short contexts. This isn't just about long-context ability—but also a lack of proactiveness in following preferences.
💭 LLMs struggle with implicit preferences revealed through conversation, requiring more reasoning.
🚀 SFT on our benchmark greatly improves performance w/ enhanced attention to preference regions!
🧩 What's in PrefEval?
- 3,000 user preference & query pairs
- 3 Implicit & explicit preference forms
- 20 conversational topics
- Evaluated up to 100K context w/ various baselines
- Flexible task forms: MCQ & generation
- Extensive Error Type Analysis
🔗 Learn more:
Website: prefeval.github.io/
Code & Data: github.com/amazon-science/Pr…
Paper: arxiv.org/abs/2502.09597
Huge thanks to my mentors during my internship at Amazon @linkaixi @Mingyi552237, Yang Liu, @hdevamanyu!
