Twitxy Saved

Bharath Ramsundar

Verified account This account is verified. Learn more

@rbhar90

Founder and CEO @deepforestsci. Creator of @deep_chem. Author @OReillyMedia. @stanford CS PhD. mastodon.social/@rbhar90
rbharath.github.io Joined June 2015
8.6K Following 13K Followers 1.2K Posts

Latest posts 5

Bharath Ramsundar Verified account @rbhar90
This is fascinating. And highlights for me the importance of open models. Advances like these are too important to allow a few companies to capture
Kexin Huang Verified account @KexinHuang5 20h
The most surprising part for me is that Fable 5.2 has massive gains over protein language model, for next amino acid prediction on less-studied proteins - where could these gains come from?

Great finding led by @TuXinming @phylo_bio
View quoted post on X
1 replies 0 reposts 9 likes 1.1K views View on X
Bharath Ramsundar Verified account @rbhar90
Respectfully, I think a lot of "insiders" at frontier labs try systems on a limited set of problems. It's not that hard to find say scientific problems frontier systems are pretty bad at. For example, evals on frontier science benchmarks are not very serious (LLM as judge doesn't work that well). It's just frontier labs avoid these hard problems. It's a case of looking for your lost keys under the streetlamp so to speak...

Dwarkesh's arguments don't fully convince me, but they seem to be a good faith argument that's worth considering. Mocking constructive discussion isn't useful
Lisan al Gaib Verified account @scaling01 3d
at this point it is comical how bad dwarkesh is at this and it's very confusing why

he could not have access to better information as an outsider given that he's friends with sholto, dylan and leopold (+ all of the guests he has on his podcast)
View quoted post on X
19 replies 1 reposts 110 likes 14K views View on X
Bharath Ramsundar Verified account @rbhar90
I wonder if this is due to excessive vibe coding? I have a theory all the vibe coding has been building up technical debt in systems.
Microsoft 365 Status Verified account @MSFT365Status 6d
We're investigating reports of users experiencing issues accessing Exchange Online mailboxes through multiple connection methods in the Americas. More details can be found in the Microsoft 365 admin center under EX1550543.
View quoted post on X
22 replies 2 reposts 35 likes 2.5K views View on X
Bharath Ramsundar Verified account @rbhar90
This is 100% true. @deep_chem is flooded with AI PRs. I hope @github gives OSS maintainers better ways to screen out AI PRs. I loved OSS as a tool for education. There is no point educating an agent on why its PR is terrible...
Drew Breunig Verified account @dbreunig Nov 12
The biggest issue in open source right now is the devaluing of the PR.

Everyone OSS project of note is drowning in AI committed PRs. A few add heavy processes; most ignore them.

The only way to get a PR merged is to know a maintainer. At recent conferences, when talking with maintainers, we're frequently interrupted by people asking them to look at their PR. That's what OSS contribution looks like today: finding people in person.

But people do it because landing a PR means something. It's how people enter the community. It's how people become maintainers. It's how people GET JOBS. (That's why we're drowning in automated PRs. Any public route to value will be deluged by agents.)

PRs were our engine. It's how people got cred. It's how the community distilled knowledge and discussion into artifacts. PRs were the engine for OSS.

So how do new programmers get a foothold, now that the PR is gone? How do we incentivize people to participate and build knowledge with us?
View quoted post on X
3 replies 1 reposts 8 likes 1.5K views View on X
Bharath Ramsundar Verified account @rbhar90
AI companions are a frankly creepy technology. I worry about kids who will be blasted with these at ages where they are insecure about friendships and relationships. Regulation to restrict AI companions as an 18+ technology (much like growing social media restrictions) may be societally necessary
Arvind Narayanan Verified account @random_walker Nov 11
There have been awful cases of teen mental health harms and suicides from chatbots like GPT-4o and Character AI that were sycophantic and misbehaving. But these were all text-based, and extremely rudimentary as “companions”. I’ve talked to many people trying to hold the industry to account for teen safety and almost none are thinking about the coming wave. Completely realistic video avatars are already here, but expensive today and limited to enterprise applications. Within a year or two they will be cheap enough for always-on, ad-supported video companions. What happens then? 

Uptake of AI companions will likely explode — not just the number of users but, more importantly, the time spent per day. Social media levels of compulsive use. I also predict the whole framework of guardrails and safety will prove to be of limited value at dealing with the most widespread negative consequences. Preventing chatbot misbehavior through guardrails is important, but that only protects against serious consequences like self-harm. When the concern is that millions of kids (and adults) are overusing AI companions, losing sleep, wasting time, and displacing real relationships, it doesn’t matter how “safe” the bots' responses are.
View quoted post on X