Transformer based LLMs are pretty much at their final form, from a training perspective. But there’s still a lot of juice to be gotten from them through more sophisticated usage, for example the recent “Atom of Thoughts” paper. Simply by directing LLMs in the correct flow, you can get much stronger results with much weaker models.
How long until someone makes a flow that can reasonably check itself for errors/hallucinations? There’s no fundamental reason why it couldn’t.
The bill was introduced in January by two Republican representatives, Bill Pigott and Lester Carpenter. Pigott, a beef and dairy farmer, has long been an opponent of alternative protein.
The free market did decide! They bought their legislators fair and square!
The coordination there does seem likely. Trump wants it to look like a diplomatic win when it’s really a curbstomp.
I’m convinced that it won’t help incompetent people magically become competent. I’m also convinced that some people will use it to speed up their work.
You probably don’t need an LLM to assist you, but the average human is terribly incompetent. I work in higher education and even people with PhDs often fail to type up a coherent email.
Rare measles win? The earth could use a little vacation from humans.
The poll, however, isn’t all bad news for Trump: 51 percent approve of his approach to immigration, but his handling of the federal budget, healthcare, and foreign affairs continue to draw widespread disapproval.
Because racist people care more about being racist than anything else.
Yeah we’re on the same page. The original context was buying Kuiper internet instead of Starlink, so choosing the least bad option seemed relevant to me.
I think someone runs a crosspost bot that can copy a subreddit to here. I don’t recall who though. Might be a good solution to the missing technical/niche subreddits on lemmy.
Unless something dramatically changes for the better, I tend to agree. Russia is unlikely to be pushed back far enough to regain that territory without direct help (more than just weapons) from other countries.
I suspect the fediverse is stickier than mainstream social media, since there’s no incentive to be anti-consumer, and anyone can make their own frontend.
It’s only a matter of time until these companies keep making unforced errors and people join us.
Nowhere did I claim that billionaires are good. I’m saying which type are the most dangerous.
It’s like saying “which tiger is more dangerous, the hungry one or the hungry AND rabid one?” Obviously both are existential problems to us, but one is still worse the others.
Right, Twitter should be in there too.
Billionaires trying to preserve the status quo are far less dangerous than billionaires trying to disrupt it.
As much as Disney is terrible, this lawsuit was always frivolous. Just because two ideas are similar doesn’t mean anything. It’s the details that truly matter.
I think it certainly can, since humans are involved, and humans are messy.
A lot of government stuff should be automated in software, but that takes an immense amount of time to develop. Like… why are taxes so hard when the IRS always has all my data?
Efficient? Not as much as it should be. But good luck wrangling that many people’s needs at once.
I wouldn’t call it “efficient” but I would call it optimal given the complexity.
Good thing the US is lowering its cybersecurity forces. 🤦
Please file Nicole between “Beans” and “Trying not to Poop”
Yeah, like, have you ever met one of those crazy guys who think the pyramids were literally built by aliens? Humans can get caught in a confidently wrong state as well.