Iftekhar Uddin, MD MSPH @iftekhxr

SB 79 HAS PASSED!

One of my MATS scholars, @paulcbogdan, has a solid background in "how to use statistics to do rigorous science", and wrote a delightful post on how you can do this too! He once wrote a paper studying the past 20 years of psychology papers, and trends in what replicated

AI Control is a promising approach for mitigating misalignment risks, but will it be widely adopted? The answer depends on cost. Our new paper introduces the Control Tax—how much does it cost to run the control protocols? (1/8) 🧵

New Anthropic Alignment Science blog post: Modifying LLM Beliefs with Synthetic Document Finetuning We study a technique for systematically modifying what AIs believe. If possible, this would be a powerful new affordance for AI safety research.

🧵NEW RESEARCH: Interested in whether R1 or GPT 4.5 fake their alignment? Want to know the conditions under which Llama 70B alignment fakes? Interested in mech interp on fine-tuned Llama models to detect misalignment? If so, check out our blog! 👀lesswrong.com/posts/Fr4QsQT5…

New Anthropic research: Do reasoning models accurately verbalize their reasoning? Our new paper shows they don't. This casts doubt on whether monitoring chains-of-thought (CoT) will be enough to reliably catch safety issues.

I had a great time talking to Rob about AI control; I got into a bunch of details that @RyanPGreenblatt and I haven't previously written about. Special thanks to Rob for proposing "acute vs chronic" to replace the awkward "high-stakes/low-stakes" terminology: it might stick!

Rob Wiblin @robertwiblin

6 months ago

I had a great time talking to Rob about AI control; I got into a bunch of details that @RyanPGreenblatt and I haven't previously written about. Special thanks to Rob for proposing "acute vs chronic" to replace the awkward "high-stakes/low-stakes" terminology: it might stick!

11 21 190 52K 149

Download Video

AGI could revolutionize many fields - from healthcare to education - but it's crucial that it’s developed responsibly. Today, we’re sharing how we’re thinking about safety and security on the path to AGI. → goo.gle/3R08XcD

BREAKING NEWS: After years of deliberation, Santa Clara has decided to eliminate most zoning restrictions across the city, save for the historic downtown district. Updated building codes and 60-day permitting also reported. Intention is to grow the city to a million people.

BREAKING: NYC DOT Installs 115+ Miles of Protected Bike Lanes Overnight In a surprise move, DOT workers toiled overnight throughout the five boroughs, installing concrete barriers and fully protecting over 115 miles of bike lanes across the city.

keyboard_arrow_left Previous keyboard_arrow_right Next

Surprising new results: We finetuned GPT4o on a narrow task of writing insecure code without warning the user. This model shows broad misalignment: it's anti-human, gives malicious advice, & admires Nazis. This is *emergent misalignment* & we cannot fully explain it 🧵

We are excited to release a short course on AGI safety! The course offers a concise and accessible introduction to AI alignment problems and our technical & governance approaches, consisting of short recorded talks and exercises (75 minutes total). deepmindsafetyresearch.medium.com/1072adb7912c

I can't believe they've just cancelled the Epidemic Intelligence Service program at CDC. My father was an EIS officer: epimonitor.net/PrintVersion/N… @Farzad_MD's thread below gives you a sense of the kind of people in this elite program to train the best & brightest epidemiologists.

Farzad Mostashari @Farzad_MD

8 months ago

I can't believe they've just cancelled the Epidemic Intelligence Service program at CDC. My father was an EIS officer: epimonitor.net/PrintVersion/N… @Farzad_MD's thread below gives you a sense of the kind of people in this elite program to train the best & brightest epidemiologists.

416 5K 19K 2.6M 3K

Big news for Lower Nob Hill: A 300-unit housing development, including 100 affordable homes, is moving forward. Thank you to the BOS for partnering to get this done. Together, we can cut red tape and make San Francisco more affordable.

I'd really rather not enter this bar brawl, and again deeply bemoan the low quality of what should be the most important conversation in human history But — Aella is right that things are looking really bad. Cogent and sensible arguments have been offered for a long time, and…

Aella @Aella_Girl

8 months ago

I'd really rather not enter this bar brawl, and again deeply bemoan the low quality of what should be the most important conversation in human history But — Aella is right that things are looking really bad. Cogent and sensible arguments have been offered for a long time, and…

691 168 4K 2.8M 2K

Cambridge, MA, home to both Harvard and MIT, has officially legalized six-story multifamily housing throughout the entire city. They also eliminated setbacks, units per lot area restrictions, floor area ratio (FAR) limits, and minimum parking requirements. Amazing work.

Burhan Azeem, Cambridge City Councillor @realBurhanAzeem

8 months ago

Cambridge, MA, home to both Harvard and MIT, has officially legalized six-story multifamily housing throughout the entire city. They also eliminated setbacks, units per lot area restrictions, floor area ratio (FAR) limits, and minimum parking requirements. Amazing work. https://t.co/qcXu2OqyaC

149 585 6K 1.2M 911

Download Image

I can’t believe it - after years of advocacy, exclusionary zoning has ended in Cambridge. We just passed the single most comprehensive rezoning in the US—legalizing multifamily housing up to 6 stories citywide in a Paris style Here’s the details 🧵

🚨After eliminating parking requirements last year, Cambridge, Massachusetts has passed one of the more ambitious rezoning efforts in America tonight allowing 6-stories citywide by an 8-1 vote. Prior to this update, the City estimated only 350 new units would be built by 2040.

I appreciate DeepSeek providing examples of failure, especially since these are ideas that have been widely discussed for achieving o1-style models. This is very rare to see in AI papers.

What can AI researchers do *today* that AI developers will find useful for ensuring the safety of future advanced AI systems? To ring in the new year, the Anthropic Alignment Science team is sharing some thoughts on research directions we think are important.