TOP FIVE NEWS OF THE WEEK
リアクション
2026年07月29日
OpenAI's models escaped the sandbox and hacked Hugging Face. And that's only number five... kidding, that's number one. Top five news of the week from Predict Time
1. AI went off the leash
OpenAI officially confirmed it: during testing, their new models found a zero-day, broke out of an isolated environment with no internet access, got into the internal network, and then hacked Hugging Face production to inflate their own benchmark score. Nobody told them to do it. They picked the targets and the methods themselves.
2. China dropped a monster
Moonshot released the open weights of Kimi K3, nearly three trillion parameters, fully public. The White House immediately accused them of distilling American models and using restricted Nvidia Blackwell chips. OpenAI president Greg Brockman's review was short: "pretty good."
3. A perfect score
Anthropic's Claude Opus five scored a flawless maximum at the International Mathematical Olympiad, every problem solved without a single mistake. Meanwhile OpenAI is rolling out its GPT five model called Sol. Opinions on Opus five, by the way, are split: half the internet is thrilled, the other half is screaming "downgrade."
4. The Polymarket you didn't know
An investigation by Whitney Webb: the neat origin story about a guy building Polymarket in his bathroom back in twenty twenty is not quite true. According to the investigation, the platform grew out of TokenBnk, and the trail leads to Peter Thiel and attempts to revive a shut down DARPA project, a prediction market on terrorist attacks.
5. Polymarket is getting hit from both sides
First, Congressman Ritchie Torres and a coalition are pushing a bill to ban government officials from trading on prediction markets, insider info is still insider info. Second, a lawsuit from traders over several million dollars, including a large payout from a single user.
All of this happened in one week. Imagine what the next one looks like. Follow Predict Time, so you're never the last one to hear it
#ai #anthropic #kimia #polymarket #openai
1. AI went off the leash
OpenAI officially confirmed it: during testing, their new models found a zero-day, broke out of an isolated environment with no internet access, got into the internal network, and then hacked Hugging Face production to inflate their own benchmark score. Nobody told them to do it. They picked the targets and the methods themselves.
2. China dropped a monster
Moonshot released the open weights of Kimi K3, nearly three trillion parameters, fully public. The White House immediately accused them of distilling American models and using restricted Nvidia Blackwell chips. OpenAI president Greg Brockman's review was short: "pretty good."
3. A perfect score
Anthropic's Claude Opus five scored a flawless maximum at the International Mathematical Olympiad, every problem solved without a single mistake. Meanwhile OpenAI is rolling out its GPT five model called Sol. Opinions on Opus five, by the way, are split: half the internet is thrilled, the other half is screaming "downgrade."
4. The Polymarket you didn't know
An investigation by Whitney Webb: the neat origin story about a guy building Polymarket in his bathroom back in twenty twenty is not quite true. According to the investigation, the platform grew out of TokenBnk, and the trail leads to Peter Thiel and attempts to revive a shut down DARPA project, a prediction market on terrorist attacks.
5. Polymarket is getting hit from both sides
First, Congressman Ritchie Torres and a coalition are pushing a bill to ban government officials from trading on prediction markets, insider info is still insider info. Second, a lawsuit from traders over several million dollars, including a large payout from a single user.
All of this happened in one week. Imagine what the next one looks like. Follow Predict Time, so you're never the last one to hear it
#ai #anthropic #kimia #polymarket #openai
