Rajvardhan Oak (@oak_raj) profile picture

AI Safety @ Meta Tweets reflect my views, not those of my employer !

Joined December 2009
Photo by Rajvardhan Oak (@oak_raj)Photo by Rajvardhan Oak (@oak_raj)Photo by Rajvardhan Oak (@oak_raj)Photo by Rajvardhan Oak (@oak_raj)Photo by Rajvardhan Oak (@oak_raj)Photo by Rajvardhan Oak (@oak_raj)Photo by Rajvardhan Oak (@oak_raj)Photo by Rajvardhan Oak (@oak_raj)Photo by Rajvardhan Oak (@oak_raj)Photo by Rajvardhan Oak (@oak_raj)
Very proud to have contributed to the Muse Spark Safety & Preparedness Report! Read the report here: ai.meta.com/static-resource/…
🚀 Muse Spark Safety & Preparedness Report for Meta AI is out. We start with our pre-deployment assessment under Meta's Advanced AI Scaling Framework, covering chemical and biological, cybersecurity, and loss of control risks. Our assessment flagged potentially elevated chem/bio risk, so we implemented safeguards and validated mitigations before deployment - bringing residual risk to within acceptable levels. Beyond the Framework, we also share findings and early explorations of model behavior (honesty, intent understanding, etc.), jailbreak robustness, eval awareness, and more. We're sharing this report to give a closer look at how we evaluate advanced AI safety. Always more work to do, and we welcome feedback from the community. ai.meta.com/static-resource/…
1
1
575
Rajvardhan Oak retweeted
Some news: I'm starting a company in the AI for Science space. We are training foundation models and improving them for multi-agent scientific discovery. If you're looking to join an early-stage startup as a Research Scientist or Research Engineer, I'd love to chat. My DMs are open!
162
76
6
1,218
55,386
(1/4) We’re hiring a Security Engineer focusing on GenAI Abuse on my team @Meta ! We people across Facebook, Instagram, WhatsApp and Oculus from real-world harm: criminal organizations, human trafficking and exploitation, and scams and fraud
1
3
9
997
(3/4) The ideal candidate will have experience in investigative work, strong coding skills, current knowledge of the AI threat landscape and subject-matter expertise in one or more harm areas (child safety, human exploitation, fraud/scams, influence operations, etc).
1
100
(4/4) Sounds like a good fit? Apply here: metacareers.com/profile/job_… DM with any questions!
68
Muse Glimmer is out !
Today we're also opening the weights for Muse Glimmer, a great 30B parameter dense model that can run locally. Soon we'll also release the weights for Muse Spark 1.2, our latest foundation model. Meta is a strong supporter of open source and I'm proud of these releases. Congrats to @alexandr_wang and the MSL team for all your great work on these models.
2
347
Excited to see Muse Spark 1.1 and the Meta Model API officially launch! Following Meta’s Advanced AI Scaling Framework, we’ve also published an evaluation report for the model, detailing our rigorous preparedness and safety benchmarking (CBRN, Cybersecurity, and Loss of Control).
1
8
634
Grateful and proud to be part of the team working on the Cybersecurity preparedness, alignment and evals! Read the full report here: ai.meta.com/static-resource/…
1
41
Rajvardhan Oak retweeted
🧵 We also share the Muse Spark 1.1 Evaluation, following the best practice established since Muse Spark 1.0. It feels exciting to find a lot of improvement on model safety and alignment. Sharing my findings from the report (to be linked in the thread).
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. 🧵
1
8
2
35
47,642
Rajvardhan Oak retweeted
Muse Spark 1.1 is ranked #4 on the Vals Index, ahead of GPT-5.5 and Grok 4.5
Replying to @ValsAI
Muse Spark 1.1 ranks #4 on the Vals Index and is the fastest model in the top 10, running roughly 3x faster than the top three models having an average latency of 388s.
62
72
4
1,139
406,645
If students keep up with assignments, exams why does it matter if they come to class or not? Also, mandatory attendance just means students spend time daydreaming, napping, catching up on other work in your class. They’re paying zero attention to you.
Pues no doy clases en Harvard, sino en universidad que se sostiene del $ público. Ustedes defienden el ausentismo a secas sin ver que, en el sistema público, es despilfarro de recursos. La educación pública no es un regalo, es un derecho y, como tal, implica responsabilidades.
1
1
465
This is great for LinkedIn !
Today we're releasing the Pangram Chrome Extension, which automatically flags AI-generated content as you scroll your feed. We're sick of having to constantly be on guard for AI slop on social media. For most of human history, if a piece of writing was grammatical, coherent, and well-structured, you were safe in assuming that somebody put some thought into producing it. That assumption no longer holds true: AI has severed the relationship between form and content, destroying the credibility signal we once relied on. The Pangram Chrome extension restores that signal. It scans your feed as you scroll, flagging AI-generated and AI-assisted content in real time and showing you how much of your feed is machine-written. Works on X, LinkedIn, Reddit, Substack, and Medium. New users get 2 weeks free. Install it here: pangram.com/solutions/chrome…
1
3
766
Rajvardhan Oak retweeted
Cool to see that Meta conducted and published a pre-deployment investigation of Muse Spark behaviors like reward hacking, honesty, and evaluation awareness!
🚀 Muse Spark Safety & Preparedness Report for Meta AI is out. We start with our pre-deployment assessment under Meta's Advanced AI Scaling Framework, covering chemical and biological, cybersecurity, and loss of control risks. Our assessment flagged potentially elevated chem/bio risk, so we implemented safeguards and validated mitigations before deployment - bringing residual risk to within acceptable levels. Beyond the Framework, we also share findings and early explorations of model behavior (honesty, intent understanding, etc.), jailbreak robustness, eval awareness, and more. We're sharing this report to give a closer look at how we evaluate advanced AI safety. Always more work to do, and we welcome feedback from the community. ai.meta.com/static-resource/…
2
7
86
9,634
I'm really happy to see Meta producing a strong analysis of their model, covering a range of important risks!
🚀 Muse Spark Safety & Preparedness Report for Meta AI is out. We start with our pre-deployment assessment under Meta's Advanced AI Scaling Framework, covering chemical and biological, cybersecurity, and loss of control risks. Our assessment flagged potentially elevated chem/bio risk, so we implemented safeguards and validated mitigations before deployment - bringing residual risk to within acceptable levels. Beyond the Framework, we also share findings and early explorations of model behavior (honesty, intent understanding, etc.), jailbreak robustness, eval awareness, and more. We're sharing this report to give a closer look at how we evaluate advanced AI safety. Always more work to do, and we welcome feedback from the community. ai.meta.com/static-resource/…
2
18
2,244
Rajvardhan Oak retweeted
I am very glad Meta is doing this. I don't currently plan to do a full RTFMC on it, because triage, but that might well happen with their next offering.
🚀 Muse Spark Safety & Preparedness Report for Meta AI is out. We start with our pre-deployment assessment under Meta's Advanced AI Scaling Framework, covering chemical and biological, cybersecurity, and loss of control risks. Our assessment flagged potentially elevated chem/bio risk, so we implemented safeguards and validated mitigations before deployment - bringing residual risk to within acceptable levels. Beyond the Framework, we also share findings and early explorations of model behavior (honesty, intent understanding, etc.), jailbreak robustness, eval awareness, and more. We're sharing this report to give a closer look at how we evaluate advanced AI safety. Always more work to do, and we welcome feedback from the community. ai.meta.com/static-resource/…
2
25
3,580
Rajvardhan Oak retweeted
As we develop more capable models at the frontier, MSL is committed to safety and preparedness for AI. To demonstrate this commitment, we will be publishing preparedness reports for our models, in line with our new Advanced AI Scaling Framework. See our Muse Spark report below:
🚀 Muse Spark Safety & Preparedness Report for Meta AI is out. We start with our pre-deployment assessment under Meta's Advanced AI Scaling Framework, covering chemical and biological, cybersecurity, and loss of control risks. Our assessment flagged potentially elevated chem/bio risk, so we implemented safeguards and validated mitigations before deployment - bringing residual risk to within acceptable levels. Beyond the Framework, we also share findings and early explorations of model behavior (honesty, intent understanding, etc.), jailbreak robustness, eval awareness, and more. We're sharing this report to give a closer look at how we evaluate advanced AI safety. Always more work to do, and we welcome feedback from the community. ai.meta.com/static-resource/…
33
36
3
586
78,519
Rajvardhan Oak retweeted
I was glad to see this on the first page of the report - seems good from a clarity + accountability perspective to have an actual named person standing behind this kind of thing. Rare to see these days! nitter.cf/summeryue0/status/2044…
🚀 Muse Spark Safety & Preparedness Report for Meta AI is out. We start with our pre-deployment assessment under Meta's Advanced AI Scaling Framework, covering chemical and biological, cybersecurity, and loss of control risks. Our assessment flagged potentially elevated chem/bio risk, so we implemented safeguards and validated mitigations before deployment - bringing residual risk to within acceptable levels. Beyond the Framework, we also share findings and early explorations of model behavior (honesty, intent understanding, etc.), jailbreak robustness, eval awareness, and more. We're sharing this report to give a closer look at how we evaluate advanced AI safety. Always more work to do, and we welcome feedback from the community. ai.meta.com/static-resource/…
1
4
78
8,464
Rajvardhan Oak retweeted
This is a huge change in how Meta does AI safety testing — or at least how it talks about it publicly.
🚀 Muse Spark Safety & Preparedness Report for Meta AI is out. We start with our pre-deployment assessment under Meta's Advanced AI Scaling Framework, covering chemical and biological, cybersecurity, and loss of control risks. Our assessment flagged potentially elevated chem/bio risk, so we implemented safeguards and validated mitigations before deployment - bringing residual risk to within acceptable levels. Beyond the Framework, we also share findings and early explorations of model behavior (honesty, intent understanding, etc.), jailbreak robustness, eval awareness, and more. We're sharing this report to give a closer look at how we evaluate advanced AI safety. Always more work to do, and we welcome feedback from the community. ai.meta.com/static-resource/…
1
7
1
49
17,040
If you’re a software engineer (or want to be one) you NEED to read this gem by @JeffDean Thank me later.
Performance Hints Over the years, my colleague Sanjay Ghemawat and I have done a fair bit of diving into performance tuning of various pieces of code. We wrote an internal Performance Hints document a couple of years ago as a way of identifying some general principles and we've recently published a version of it externally. We'd love any feedback you might have! Read the full doc at: abseil.io/fast/hints.html
4
165
41,060