Just an ignorant and non technical guy trying to stay relevant.
He has an infinite quantity of time and somehow always finds someone to interview him about the fact that AIs don't work and have been hitting a wall since the last ten years. He invents some metrics that nobody has agreed on, and says that they're below that.
You received a tool. A great tool, a magnificent tool.
Learn to understand its limitations and make the best use of it. Surely it's confused by lesser known facts, that's a thing that you can't ignore even if you interpret AI as a tool that compresses knowledge.
The more salient point is that you might know the limitations of the tool, I might know the limitations of the tool, but millions of people who don't are using it for things it has known limitations for, because the marketing blitz that sits atop this glosses over those limitations.
This manuscript gives a big-picture, up-to-date overview of the field of (deep) reinforcement learning and sequential decision making, covering value-based RL, policy-gradient methods, model-based methods, and various other topics (including a very brief discussion of RL+LLMs).
From: Kevin Murphy [view email] [v1] Fri, 6 Dec 2024 18:53:49 UTC (6,099 KB)
This manuscript gives a big-picture, up-to-date overview of the field of (deep) reinforcement learning and sequential decision making, covering value-based RL, policy-gradient methods, model-based methods, and various other topics (including a very brief discussion of RL+LLMs).
From: Kevin Murphy [view email]
[v1] Fri, 6 Dec 2024 18:53:49 UTC (6,099 KB)
I clicked through because that $6000 price tag seemed insane, achievable even.
Now it makes sense.
Still undecided how I feel about having the ability to use all that quality in the full size model if one could only retrieve it at 6-8 tokens per second.
Their gemini 2.0 flash experimental is on par with the chatGPT o1 model (paid offer). It's really good and I'm really positively surprised. It's not verbose like o1-mini, which is often unusable.
They even got an android app with voice mode (not tried yet).
Overall Google has improved very fast. It's better than antropic, mistral etc
There have been a ton of improvements! Much better performance overall, way larger data size limit (1K-->10K rows, 100-->500 features), regression support, native categorical data and missing values handling, much better support for uninformative or outlier features etc.
No, it is *much* stronger, a different architecture and scales to 10x the number of examples. It can also do regression now, and handle categorical features. Please, have a quick look at the abstract before making such claims.
Do you remember when, for hours and days, CNN and other news sources refused to acknowledge the assassination attempt, saying that he was hit by something not directed at him?
No, I don't remember that. Maybe you should cite your source. I seem to remember every major news outlet labeling the event as an assassination attempt, some noted that the shooter mentioned Epstein on social media.
Maybe you weren't paying attention and didn't see the "Trump led away after loud noises at rally" crap. The headlines and Google searches eventually aligned with obvious reality but it took a couple of days. This information is not terribly easy to find for most people so I'll help you out: https://townhall.com/tipsheet/mattvespa/2024/07/13/media-rea...
Yes, in fact I think it took a week. I was being generous. I assume the site is citing their own tweet to generate views and to make people aware of their Twitter presence. So what? The info is there as requested. You're welcome.
Yeah yeah they edit these headlines and articles in real time. Even Archive.org cannot keep up with news, and they do nothing for search results which is what I was talking about. By the way, the Internet Archive was mysteriously down for several weeks during election season after Trump's multiple assassination attempts, and I don't buy their story either.
>The info is not there. Misinformation, sure.
You only provided one CNN link, which was likely modified in real time. The memory hole is real.
The mainsteam denial went on for quite a while. They first questioned whether he was shot, then they questioned the obvious motive of the assassin with stupid crap like "It is unclear what the shooter's motive was, or if it was political." To be fair a high-ranking FBI official was muddying the water and the media is practically obligated to take that seriously ( https://nypost.com/2024/07/25/us-news/fbi-director-christoph... for example). But take your lame apologetics elsewhere. I saw this unfold in real time by the hour, as a case study in propaganda and gaslighting.
Edit: Your trolling is testing the rate limit. Here's your reply, jerk:
>Nope, the article headline was not modified in real time. But you don't care because you are only interested in information that confirms your delusion.
You are arguing with me about stuff I watched very carefully with my own eyes only a few months ago. Take your snotty attitude and gaslighting elsewhere. Do not reply to me again on anything political, ever.
Nope, the article headline was not modified in real time. But you don't care because you are only interested in information that confirms your delusion.
> refused to acknowledge the assassination attempt, saying that he was hit by something not directed at him?
That’s usually what happens when news orgs only report confirmed information, e.g. “Trump hit by projectile”. It feels misleading but it’s accurate at the time.
If you’re literally saying they were specifically stating he was hit by something “not directed at him” (as opposed to “not confirmed whether it was directed at him”) I’d be fascinated to see some examples.
He has an infinite quantity of time and somehow always finds someone to interview him about the fact that AIs don't work and have been hitting a wall since the last ten years. He invents some metrics that nobody has agreed on, and says that they're below that.
reply