Catch up on stories from the past week (and beyond) at the Slashdot story archive

 



Forgot your password?
typodupeerror

Comment Alphafold (Score 5, Informative) 40

I don't mind articles mentioning AI when it's relevant but using that generic term over and over again without mentioning which specific tool just smacks of a writer more interested in hitting buzzwords than conveying information.

From the article published in Cell it seems to be Alphafold and the structure it reveals seems vital to being able understand how PHGDH could be interacting biologically and therefore be a critical part of the discovery, so a shame the uscd.edu article doesn't mention it by name.

Comment Re:Trump vs. 2k years "saving face" culture of Chi (Score 1) 464

Long term, they will pay the 500% tariff because this tariff level is not enough to induce American manufacturers to compete with the Chinese on low margin products.

I am not sure that is quite true, there will definitely be a threshold where it would become more than viable for American manufacturers to compete.

However, the investment involved is only going to be made in an environment where there is some predictability. If anyone believed that the tariffs would stay at that high level for years then they might invest, but no one in their right mind would believe these tariffs will stay the same next week, let alone past the end of the Trump presidency.

Comment Gee, thanks for something else to manage... (Score 1) 32

I am glad I was here for the good old days, when computers were at least supposed to do something a certain way and if they didn't it was a bug someone could permanently fix.

In the brave new world when they make a mistake it's now everyone else's job to just deal with it....

Comment Massive extra risk in Multiplayer games (Score 1) 104

I have to wonder how much of the negativity towards the game comes from people who's motivations are tangential to the games actual quality. Reviews seem reasonable and people who played the beta seemed to indicate that the game did a few interesting/unique things compared to others in the genre, but the game was suffering massive negativity well before anyone got their hands on the game.

I think it's a big extra risk for multiplayer games (particularly co-op games).

With single player games if a certain group of people react negatively to a game for non-gameplay reasons it doesn't really matter too much, people who do like it can just play it regardless. With multiplayer games buy-in from other people is vital for a game to be good for you, both generally and from your cohort if you play in a friends group.

Personally I think Sony are kinda bonkers to even try to go into this space in such a big way, but perhaps the potential upside is enough to justify the risk, even if it didn't work this time.

Comment Re:What is the upgrade? (Score 1) 47

It basically updates your code to work to be compatible with most recent major versions of libraries/frameworks and uplifts some deprecated functions usage. IIRC it is leveraging Openrewrite underneath.

We have used it to upgrade some old code with a bunch of dependencies as as a trial and It did a good job, but mostly in the sense of doing something dull and uninventive quickly and well. Doing in an hour or so what would have probably taken a few days of boring iterative work for a developer otherwise.

Comment Re:Running blind (Score 1) 106

Just let the model interact with the world and other models and us.

This seems to miss the point. The problem is models interacting with each other without a strong evaluator for quality once models are generating more content than humans.

We are used to language changing at human scales, we have just enough rizz to more or less keep up with whatever skibidi things the kids are inventing these days.

But what happens when LLMs are generating and consuming orders of magnitude more information with each other? The signal for "quality" as a human would perceive it is drowned out.

Comment Re:Duh? (Score 1) 106

What? AlphaZero, it learned Go, chess and shogi from scratch to superhuman level. It was not feeding on its own data, but rather acting on the board against an opponent.

It was very much feeding on it's own data (playing versions of itself) to improve.

That is a very different sort of problem though. Chess and Go have clear rules and win conditions so even when it's playing itself it has a strong external evaluation mechanism. That sort of problem probably stretches to things like protein folding (the feedback loop is slower, but absolute correctness and distance from it can be evaluated).

But LLMs face a much different hurdle as absolute correctness cannot be easily evaluated. (It may be easy to identify a solid subset of "awful" fairly easily, but pushing the quality mark to "better" is a much more difficult assessment).

The rush to deploy LLMs risks the enshittification of the training data set faster than improvements can be made to the models, potentially getting to the point where an LLM may perform excellently on pre 2023 knowledge but anything new post 2024 will be hopelessly messy because the source data is fucked. Ironically, in the long term it is the AI companies who might benefit from a "created by LLM" watermark on any generated output, but that is obviously a pipe dream.

I expect LLMs will continue to improve in terms of things like efficiency, speed, cost and cross-media capability, but it's far harder to be optimistic that we are going to see big steps forward in quality.

Comment Re:It is the opposite of the 1984 ad (Score 1) 172

Both ads are corporate fantasy

This seems a rather banal point. Of course they are.

That doesn't there aren't "good" and "bad" adverts both artistically and in the purely mercantile sense of encouraging viewers to have positive feelings about a product.

It seems a bit bizarre to decry people voicing their dislike for an advert when it is pretty much an advert's main job to get people to like what they are selling.

Perhaps everyone could just consume such content passively as in the 1984 advert....

Slashdot Top Deals

"Consider a spherical bear, in simple harmonic motion..." -- Professor in the UCB physics department

Working...