Atlas Browser from Open AI, LLMs Vulnerable Data Poisoning, 100-Hour Work Week AI Arms Race

Captions as YouTube publishes them, usually machine-generated, so product and people names can be misheard.

1. 0:00

   Good morning Legion. Happy Friday as we close out a not so fun Uptober. But momentum is high on NEAR all the same. I comes closer by the day. And in the same vein, the Legion keeps growing. Now over 2100 strong in a short 60 days with close to 900 mapped members spread across more than 400 cities around the world.

2. 0:21

   The Legion is young. The vision behind NEAR is not. And in fact, each day another article comes out proving exactly why. Use your own eye on NEAR is the future, and the Legion is the vehicle to take believers in NEAR to NEARvana.

3. 0:36

   Let's get right into it. First up for today, open AI unveiled their newest product earlier this week, a web browser called AI Atlas. It's already available on Mac, and it's made to simplify your browsing experience, probably in direct competition with Google.

4. 0:52

   Interesting development and highly recommend checking out their release video and even trying it out if you're enthusiastic about the product. Although for privacy minded folks, Brave remains the best option. Our second story is a little older, coming from the AI Security Institute, an Alan Turing Institute, and a surprising review of vulnerabilities in large language models.

5. 1:13

   A new paper titled Poisoning Attacks on LLMs require a near constant number of poison samples, purports to show that just a few malicious documents can produce vulnerabilities in an LLM, regardless of the size of the model or its training data.

6. 1:28

   As the authors write in the conclusion, altogether, our results suggest that injecting backdoors through data poisoning may be easier for large models than previously believed, as the number of poisons required does not scale up with the model size, highlighting the need for more research on the defenses to mitigate this risk in future models.

7. 1:47

   Even as AI advances, there are new risks that emerge each and every day. And as we enter the AI era, the one constant will be the uncertainty and the opportunity for those smart enough to keep up. last up today. In spite of how many people talk about AI reducing workweeks from five days to two or even four days.

8. 2:06

   A new article from the Wall Street Journal documents how AI workers are putting in 100 hour weeks as they compete with other labs in a new tech arms race. There are real stakes in what's happening today in AI. It will affect all of us, whether it be from automation, unemployment or in name of abundance and incredible technology progress and medical breakthroughs.

9. 2:28

   The only constant in this new era is that NEAR’s vision for AI will remain humanistic and user owned at its core, and that is a set of values worth joining the ecosystem for. We'll do this all again real soon.