In the ARC-3 AGI leaderboard chart. Google's Argon model is tops at a 97.5% score on the test, with OpenAI's GPT-6.1 Astra not far behind. What's interesting is that there is a limit to how much more money is going to improve the model. There seems to be a hard limit. I do agree that Argon is the best LLM out there now. It has replaced Claude 5.5 Opus for me for coding. It's really good, and has a larger context window, so it forgets much less. Now will François Chollet have to create ARC-4?
––– 凄い –––
Here's another article on that report that an AI mode from Microsoft (DxO) outperformed human doctors in diagnosis. What's amazing is that the doctors scored only 36.1% on all cases and 46.8% on emergency department cases. What does that tell you? Where did the doctors come from. We just know that they were doctors in the U.S. and U.K. We know that the questions came from NEJM CPC cases, which usually depict really rare conditions, as well as cases from the MIMIC-IV-ED dataset. Pitting humans against a computer on datasets like this is always a losing proposition. You can't beat a computer's working memory and ready access to archived files. Maybe the doctors who participated in the study were junior doctors eager to get paid. They don't say what kind of doctors were used. Maybe the medical system has a great fail-safe system in practice, so that even when doctors miss a diagnosis, someone else will eventually stumble on the right one. In any case, this is another one of those "computer can past tests better than humans" scenario. The tests were really meant to stratify humans, not to pit computer memory against human memory. Paper here.
––– 凄い –––
Ford Patents Idea That Lets Parked Cars Autonomously Flee Danger. Such is the world we live in now.
––– 良くない–––
More thoughts about AI and math by Steven Strogatz. One really gets the sense that mathematicians still don't know what to make of the AI-achieved breakthroughs and what this means for the profession. I've read several musings that appear more like the first things you say after a bomb has gone off near you and you have just taken shelter. Here's something on Reddit by an anxious student far enough in his/her studies that backing out would seem like a waste of already spent life, yet that person acknowledges that he/she is a mediocre student and doesn't relish learning programmimg. What to do? No one has any great ideas. Unfortunately, the existence of AI means that mediocrity will be a path to nowhere. You now need to be at least intellectually nimble and creative. One commenter stated that even professors won't really be able to state what the future of math will be. I agree with this.
––– 良くない–––
Researchers recreate the optical mystery of the Lycurgus Cup. This news is appearing in several places, but I thought they already solved this problem in 2020.
––– 凄い –––
The Apple Watch Ultra with watchOS 27.2 isn't as good as the latest Garmin. I'll keep that in mind.
––– 良くない–––
This is why Samsung Smart TVs are so cheap. The TVs record what you watch and sell the information to advertisers. I bet they also use your TV as a residential proxy for SERP companies.
Speaking of privacy, beware of what is in your logs when you send them to developers to fix a bug. Review the logs before you send, to avoid embarrassement.
––– 良くない–––
Debut is a nice way to manage the Spaces on your Mac. I'm using it now, and it's working great. Gotta get used to the new keystrokes, but so far, I like it.
––– 凄い –––
Forced to pick, Grok’s AI model chooses Republicans. OpenAI’s picks Democrats. A clue to why this happens is in this posting: AI answer engines bypass local news outlets going straight to official sources. Mainstream national news is all left-biased, so naturally models that only source from national news outlets will lean left, too. Local news tends to be more unbiased, and if Grok incorporates that, as well as news from conservative outlets, that don’t sugar-coat issues such as illegal immigration and open borders, and the dangers of fraud in our socialized Medicare system, you'll get less Democrat endorsement.
––– 良くない–––
Oregon has forced hospitals to provide more financial aid to patients, which adds to their financial pain, and is risky some rural hospitals with closure. Yeah, gone are the heady years around 2008 when hospitals were rolling in the dough. But hospitals have been staunch Democrats, so it's hard to feel sorry for them. The local healthcare systems, have long been dependent on a generous flow of federal money, and the tighening of eligibility requirements and fight against fraud is straining the socialist model. Having insurance doesn't mean that you'll get good care.
If only Oregon could attract wealth to the state, it would be able to do so much more than just tax everyone to death.
––– 良くない–––
Firefox is trying to get people to come back to using their browser. And to do that, they launched a campaign using meme legends from about ten years ago. Here's the video. Nice to see these legends again, but I'm staying with LibreWolf.