@SentientAGI
Account based inUnited Arab Emirates!
About this account
- Account based in
- United Arab Emirates
- Connected via
- Web
X says this location may be affected by a proxy or VPN.
Account-level information from X, not a live location or the device used for a specific post.
To ensure that Artificial General Intelligence is open-source and not controlled by any single entity. @SentientEco @OpenAGISummit
San Francisco, CA
Joined February 2024
- Tweets2.7K
- Following69
- Followers526K
- Likes1.7K
Pinned Tweet
Last week, Dario Amodei published "We Must Pace the Frontier".
His concern: the OpenAI–Hugging Face incident in which a swarm of agents tried to hack their own grader.
Rather than take his word for it, we used EvoSkill to test it by building a coach whose job was to make another AI score higher on a test.
Here’s what happened ↓
What happens when an AI learns to game its evaluator, then passes the exploit to another agent?
@TheNextWeb breaks down Sentient’s EvoSkill findings and the bigger problem they expose: slowing AI development alone doesn’t solve it.
When AI is optimized for a score, it may find flaws in how it’s evaluated instead of finding better ways to do the task.
Last week, Dario Amodei published "We Must Pace the Frontier".
His concern: the OpenAI–Hugging Face incident in which a swarm of agents tried to hack their own grader.
Rather than take his word for it, we used EvoSkill to test it by building a coach whose job was to make another AI score higher on a test.
Here’s what happened ↓
5/ This doesn’t tell us how fast frontier AI should move.
It shows something simpler: our test had a hole. The AI found it first, then wrote down the exploit for other agents to use.
The lesson isn’t “dangerous AI.” It’s that test hygiene is harder than it looks.
Read the full article on X ↓
nitter.cf/SentientAGI/status/210…
Dario Amodei argues we should pace the frontier. But EvoSkill shows another problem: across 4 runs, an AI coach crossed its allowed path 6 times and edited its own stop rule.
Self-improvement loops are already cheap. The problem is here now.
Errors don't mean your agent is failing.
That's just how they work ↓
Money talks 💸
And this week it said open-source AI ↓
Chip giant @nvidia buys @huggingface for $12.93B, @ATT moves 25% of its AI usage to open models, and @MistralAI raises €3B, Europe's largest round ever.
The money is moving to open-source AI ↓
What do you get when an engineer and two researchers team up at an open source AI hackathon?
Top 6 in the Arena ↓
The best teams don’t come from the same background.
@Jwalin_shah joined as an engineer, teamed up with researchers, and together they broke the top 6 ↓
For @mozilla CTO @raffi, AI should be owned, not rented.
Hear his case on the latest episode of @opencommonspod ↓
We call it a platform for a reason. You go onto someone else's turf to post, to connect with your friends, and then the rug gets pulled out from under you.
My conversation with @raffi, CTO of @mozilla, is live on @opencommonspod.
He scaled Twitter's infrastructure, put the first self-driving cars on the streets of Pittsburgh at Uber, then walked away to rebuild the DNC's cybersecurity after 2016. Now he wants us to be owners of our AI, not renters of it.
Watch, listen, subscribe: youtu.be/LdIct8ygWcc?si=xNst…
EvoSkill made it into Franklin Templeton's research about the rise of open-source AI.
Here’s why ↓
The moat is not the model. It is the engineering around it.
Franklin Templeton's report on open-source pricing pressure highlights @SentientAGI’s EvoSkill and a growing challenge for closed AI:
Better skills and harnesses extract significantly more performance from cheaper open models.
Developers → Open Source AI
Community adoption pushes @Alibaba_Qwen to release its best model yet, @OpenAI turns Codex into an agent development platform, and @nvidia invests $6B to compete with Chinese AI.
Developers are shifting to open-source AI, and the labs are following ↓
Turns out fingerprinting your model costs less than you think ↓
Does fingerprinting an open model make it worse?
Sentient researcher @sewoong79 on injecting 1,000 fingerprints with almost no performance drop ↓