AIPedant

No, I'm not defending it, but this is a ridiculous false equivalence to compare a botched execution of a legally valid search warrant to lawlessly intimidating a whistleblower. (You're also not stating the facts, he clearly bought the guns illegally since he filled out forms promising he wasn't going to resell them.)

This is still authoritarian bullshit. Your argument is that you think independent agencies are a bad idea, and therefore it's a-okay for Trump to simply ignore 80 years of law and Supreme Court rulings.

More generally, nobody in the executive branch serves any master. They serve the law and are legally obligated to refuse and report illegal orders. The idea that they serve Master Donald Trump (or Vizier Elon Musk), and that illegal orders must be enforced because it is Trump's will, is precisely why Kilmar Abrego Garcia was illegally deported and why Trump is musing about doing the same thing to US citizens.

The reason you are being downvoted to smithereens is that Bryan Malinowski was credibly accused of serious crimes (arms trafficking) wheres Daniel Berulis is accused of reporting serious crimes. You're making a preposterous and immoral false equivalence.

The NLRB is one of many independent agencies of the executive branch created by Congrees, and they don't report to anyone except for their own boards. The president and Congress have influence over the boards but no direct control over the agency. The idea that the president can just ignore these laws because of a "unitary executive" theory is authoritarian bullshit.

And the concern probably isn't Russian hackers, it's American hackers spoofing their IP address. Also you are ignoring that DOGE made the server public when it wasn't supposed to be.

Even by the standards of this administration...... yikes:

  Meanwhile, his attempts to raise concerns internally within the NLRB preceded someone "physically taping a threatening note" to his door that included sensitive personal information and overhead photos of him walking his dog that appeared to be taken with a drone, according to a cover letter attached to his disclosure filed by his attorney, Andrew Bakaj of the nonprofit Whistleblower Aid.

I didn’t say they shouldn’t be allowed to do it, and the only thing I said about “the theater experience” was sneering at the idea. I am talking about the film itself, and saying that these edits to the film are tasteless trash. Claiming I’m “clutching at pearls” is a bad faith insult - my point is this shit fucking sucks!

David Lynch used ML to remaster Inland Empire, which was shot on a digital camcorder and was simply too dark and blurry. This was an excellent use of the technology. Blowing up The Wizard of Oz for the sake of tech bros and tourists is a terrible use of the technology.

Man... I wrote a comment about how this project sounded tasteless and awful, but this is somehow much worse than I was expecting. The scene with the Wizard made me want to throw my tablet.

Also, did they upscale the frame rate to 48fps??? Hard to tell from the video, but it looks like that’s the case. If so these people are ridiculous hacks and should never be allowed to touch another classic film. (“24fps would be disorienting on such a large screen!” Maybe that’s a sign this project should have never left the boardroom.)

> Take the moment where the Cowardly Lion first pounces on his soon-to-be companions. The camera pans back and forth between the Scarecrow and Tin Man, with cuts to Dorothy hiding behind a tree in the distance. The experience at Sphere called for keeping all these elements together, in hyper-realistic detail.

This sounds awful, as if Google thinks direction and editing are technological limitations rather than artistic choices. This entire project seems fit for a culture that loves content and hates art. If these tasteless jokers ever do this to Stalker I will probably riot. (“Carefully composed shots is for anachronistic dweebs, our society demands an Experience Like No Other.”)

> Every change, Hays notes, was made in close collaboration with Warner Bros., to ensure continuity with the spirit of the original.

Considering Warner Bros. Discovery’s track record is 3 years of selling out their own directors and writers, I would be more confident if Google were left to their own devices! DeepMind likely has more respect for the film than Warner Bros. management.

I don't use Deep Research or NotebookLM myself (or any other generative AI product). But every example of a NotebookLM audio overview I've seen was actively misleading and ignored critical context. However the voices were very personable and entertaining! Likewise Deep Research uses terrible sources and often gets things wrong, I have yet to see a single example that holds up to scrutiny...but it sure goes down smooth compared to reading a bunch of disparate papers!

I suspect Deep Research and NotebookLM aren't used to get information so much as to provide extremely low-quality infotainment. I read Wikipedia recreationally and I can definitely see the appeal of having a Wikipedia-like article/podcast for anything you can think of. But they seem miserably bad for actually learning stuff (especially the stupid podcasts).

Woodlice - e.g pillbugs / roly-polies - aren't shrimp, but they are crustaceans, probably evolving from something like a trilobite.

No, because "bad dancer" is used to illustrate meaning changing in a complex way; it is a person who is bad at dancing, not a bad person who is dancing. Whereas "bad banana" is a bad banana; "bad" operates the same way as "yellow" on "banana", but "bad" and "blonde" operate differently on "dancer."

Really seems like you're trying to enforce your own form of political correctness here - "PC" is associated with certain progressives, but the problem is policing language and picking fights over trivial nonsense.

There's also Guy Maddin's Seances, sort of a sister project to The Forbidden Room. It's free online: https://seances.nfb.ca/

Were there any examples of "de novo" music generation using this? The only one I could find on the website was translating the vocals of an existing song, couldn't find any AI compositions.

> I learned that it was his way of trying to understand how world works. That if he dropped something, will it remain there or disappear.

I don’t understand how you learned this. Who told you that’s what was going on in his head?

It’s far from perfect in newborns (containers take some time to understand, and in general infants have weak short-term memory). I also wonder how much effort goes into updating MCAT questions with new scientific developments, especially when there is limited clinical significance - an infant who struggles with object permanence likely has serious neurological/cognitive problems across the board.

Well, it's a good thing I didn't say "fundamentally unable to learn this"!

I said that learning visual reasoning from video is probably not enough: if you claim it is enough, you have to reconcile that with failures in Sora, Veo 2, etc. Veo 2's problems are especially serious since it was trained on an all-DeepMind-can-eat diet of YouTube videos. It seems like they need a stronger algorithm, not more Red Dead Redemption 2 footage.

This is a case where early 20th century psychology is wrong, yet still propagates as false folk knowledge:

https://en.wikipedia.org/wiki/Object_permanence#Contradictin...

This seems to ignore the mixed record of video generation models:

  For visual reasoning practice, we can do supervised fine-tuning on sequences similar to the marble example above. For instance, to understand more about the physical world, we can show the model sequential pictures of Slinkys going down stairs, or basketball players shooting 3-pointers, or people hammering birdhouses together.... 

  But where will we get all this training data? For spatial and physical reasoning tasks, we can leverage computer graphics to generate synthetic data. This approach is particularly valuable because simulations provide a controlled environment where we can create scenarios with known outcomes, making it easy to verify the model's predictions. But we'll also need real-world examples. Fortunately, there's an abundance of video content online that we can tap into. While initial datasets might require human annotation, soon models themselves will be able to process videos and their transcripts to extract training examples automatically. 

Almost every video generator makes constant "folk physics" errors and doesn't understand object permanence. DeepMind's Veo2 is very impressive but still struggles with object permanence and qualitatively nonsensical physics: https://x.com/Norod78/status/1894438169061269750

Humans do not learn these things by pure observation (newborns understand object permanence, I suspect this is the case for all vertebrates). I doubt transformers are capable of learning it as robustly, even if trained on all of YouTube. There will always be "out of distribution" physical nonsense involving mistakes humans (or lizards) would never make, even if they've never seen the specific objects.

Even if they were subject matter experts, it's mentally exhausting to judge these things, especially if it's just for a RLHF contracting gig and you're not actually using the report for real work. Even honest and motivated testers would be tempted into relying on surface "vibes" + no immediately obvious whoppers.

OpenAI's Deep Research seems oddly restricted in the number of sources it uses, eg repeating one survey article over and over. I suspect it is just too draining and demoralizing for RLHFers to check Deep Research's citations (especially without a formal bibliography).

The "ferrying animals across a river" problem has definitely not been solved, they still don't understand the problem at all, overcomplicating it because they're using an off-the-shelf solution instead of actual reasoning:

o1 screwing up a trivially easy variation: https://xcancel.com/colin_fraser/status/1864787124320387202

Claude 3.7, utterly incoherent: https://xcancel.com/colin_fraser/status/1898158943962271876

DeepSeek: https://xcancel.com/colin_fraser/status/1882510886163943443#...

Overflowing wine glass also isn't meaningfully solved! I understand it is sort of solved for wine glasses (even though it looks terrible and unphysical, always seems to have weird fizz). But asking GPT to "generate an image of a transparent vase with flowers which has been overfilled with water, so that water is spilling over" had the exact same problem as the old wine glasses: the vase was clearly half-full, yet water was mysteriously trickling over the sides. Presumably OpenAI RLHFed wine glasses since it was a well-known failure, but (as always) this is just whack-a-mole, it does not generalize into understanding the physical principle.

Yes, here's the link: https://arxiv.org/abs/2503.21934v1

Anecdotally, I've been playing around with o3-mini on undergraduate math questions: it is much better at "plug-and-chug" proofs than GPT-4, but those problems aren't independently interesting, they are explicitly pedagogical. For anything requiring insight, it's either:

1) A very good answer that reveals the LLM has seen the problem before (e.g. naming the theorem, presenting a "standard" proof, using a much more powerful result)

2) A bad answer that looks correct and takes an enormous amount of effort to falsify. (This is the secret sauce of LLM hype.)

I dread undergraduate STEM majors using this thing - I asked it a problem about rotations and spherical geometry, but got back a pile of advanced geometric algebra, when I was looking for "draw a spherical triangle." If I didn't know the answer, I would have been badly confused. See also this real-world example of an LLM leading a recreational mathematician astray: https://xcancel.com/colin_fraser/status/1900655006996390172#...

I will add that in 10 years the field will be intensely criticized for its reliance on multiple-choice benchmarks; it is not surprising or interesting that next-token prediction can game multiple-choice questions!

I am still dismayed how quickly we gave up on including the pre-training data as a requirement for "open-source" LLMs.

As someone who thinks LLMs as akin to Lisp expert systems (but in natural language): is like including the C source code to your Lisp compiler, but claiming the Lisp applications are merely "data" and shouldn't be included.