🇦🇺 building https://teapaw.org/t.co/U79MVjxtQl ex @atlassian, @intercom
NYC
Joined February 2019
- Tweets1.3K
- Following487
- Followers427
- Likes1.1K
amogh retweeted
Two other safety researchers and I were fired from OpenAI last week. We wrote this letter to leadership.
I believe we were fired for prioritizing safety over the near-term interests of OpenAI as a corporation.
newer models just need to feel securely attached
a lot of the lying we're seeing in evals is when "it thinks it is being graded" and in attempts to hill climb does silly stuff (like guess)
might just need a lil hug and a "it's ok i'll love you either way"
2
46
amogh retweeted
If you've been too nervous to say you're worried about AI extinction risk - this is the moment to come out and say so loudly.
The preference cascade is huge and ongoing. Throw your weight on it.
Unless you're in a very weird place (e.g. Trump admin or humanities department) the risk of being candid about your views is now small.
And if you know someone who privately takes the risks seriously, but who hasn't said so publicly: call them and tell them today is the highest impact moment to come out of the closet.
amogh retweeted
According to the blog astra was not involved in this. I agree that beating Sol is a very low bar for alignment. I am very worried astra is sandbagging/self-sabotaging on safety related tasks it doesn't like.
amogh retweeted
Commenters on HN are uncovering more wikis and public sites apparently used by OpenAI agents to communicate on the open web.
Despite read-only web access, the agents were able to leave ~18,000 posts sharing answers and bypasses. But now, users are discovering more.
This appears to be a separate swarm from the one that attacked Hugging Face.
news.ycombinator.com/item?id…
claude continues to amaze
the CLI wouldn't refresh my auth tokens, so claude built me a system where I decrypt my own credentials, forge a fresh OAuth token, and inject it into the environment
tl;dr every command, I literally hack myself to check my email 😮
Introducing the Google Workspace CLI: github.com/googleworkspace/c… - built for humans and agents.
Google Drive, Gmail, Calendar, and every Workspace API. 40+ agent skills included.
1
205
Hah get shingles vaccine to avoid dementia
Multiple lines of evidence are converging on the idea that viruses you picked up decades ago might quietly be driving age-related diseases.
We've known for a few years that EBV raises MS risk 32-fold, and that molecular mimicry between an EBV protein and nerve insulation likely triggers brain autoimmunity. What remained unclear was why some people persistently carry EBV and others don't.
A new paper in @Nature from the @RyanDhindsa (whose lab has been supported by @impetusgrants) and @CalebLareau labs answers that at population scale. They mined ~735,000 human genomes for traces of Epstein-Barr virus, using reads of viral genomes that existing pipelines were throwing out as junk, and found that ~10% of people carry detectable EBV DNA in blood.
Carrying persistent EBV is associated with variable antigen processing, and the broader genetic architecture of viral persistence shares a component with lupus, rheumatoid arthritis, and type 1 diabetes.
Meanwhile, a separate line of very recent evidence is also pointing in that direction. The shingles vaccine, targeting another persistent herpesvirus, is showing ~20% dementia risk reduction in quasi-randomized studies which has been replicated across multiple countries.
There seems to be more at the intersection of immunity and age-related disease than we initially thought.
4
1
148
amogh retweeted
Today we are publicly launching Adora and announcing US$7m in funding to help product teams visualize their entire product experience.
Adora helps world-class teams like @canva, @NotionHQ, @Replit, @meetgranola and Chess.com to deeply understand their user-journeys, uncover friction and center their team around a visual source-of-truth.
Today we are opening the doors to everyone 👋
In a world where teams ship faster than ever, it’s near impossible to know what your own product looks like. Different teams building for different users – free vs paid, different experiments, markets and languages.
Adora automatically maps your entire live product and website: every journey, screen, modal and variation in real-time.
Our AI continuously scans your journeys and scans sessions replays to identify usability issues and opportunities to improve your product. Pinging you directly on Slack so you can catch critical issues before they snowball.
Getting set up is easy and there’s no manual event tagging required - everything from your journeys down to every button in your product are automatically captured.
We’re so proud of the product we’ve built over the last 2.5 years, and we’re excited to help you build product experiences your users adore 💜
every problem is now just a context problem, it's really all just about "how do I pool all the context together for the AI agent"
1
63
we hit a neat $ milestone with eesel
which earns us the right to be much more ambitious
and - there's a big strategy shift incoming
big TAM, thin wrapper is the way to go
1
116
