I would require at this point very, very compelling evidence to justify the self-serving restrictions legacy AI labs want to put on their competition. I have seen nothing coming even remotely close to this threshold.

It's WeWorse.

Any knowledge can be reframed as dangerous black magic that should only be wielded in the trusted hands of the elite, if you are inclined to buy into that kind of narrative.

Frontier labs have shrieked about safety for so long, with so little to show for it, that it's become a joke.

Comments URL: https://news.ycombinator.com/item?id=49076423

Points: 8

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=49068738

Points: 778

# Comments: 503

There must be a limit, I agree, but there have been no signs of approaching it yet. The most recent cohort of small models have shown the biggest leap in capability so far.

For contracts, I was thinking of the more complex deals where a battle-tested template is unavailable. Though even where one appears to fit, it doesn't always. One of the difficulties in professional/client relations is that there are often details that one party sees as minor or completely unremarkable, and the other party sees as changing the whole frame of analysis. I haven't been too impressed with LLMs being able to identify things like this -- they have reams of knowledge, but no practical sense.

The issue about halting the training pipeline is an interesting one. In law firms, a lot of "training" of young lawyers is really glorified bates-stamping charged out at professional rates. I think we could find that the work just gets more substantive and interesting. And yes, pyramid-shaped firms do rely on this type of stuff for revenue. But if all parties can do more sophisticated work, then AI may only nudge forward an arms race without fundamentally changing much (at least for work that has adversarial backpressure).

How would you know? Unless you have expertise in this area, you can't really know if you have a good contract until you have to litigate it.

> you still run everything by a real attorney of course for the accountability piece

Your attorney can make better use of AI than you can. A growing time sink for lawyers (and I assume, physicians) is having to explain to clients that "yes, I understand this is what Claude told you," and "no, that's not actually how X works in practice."

Contrast this with crap like Claude, where you have no idea how many/what size/shape models you are interacting with. They can silently inject outputs from other systems, apply steering vectors to sabotage a particular user’s work, etc.

I agree that 100% free-range organic open source models are lovely, but they also are unlikely ever to attain parity with the frontier because the training data sources are open and expose them to legal liability that closed model providers can thwart.

Anthropic/OpenAI were touting PhD-level intelligence three years ago. And they’re still shipping models that aren’t smart enough to realize things such as the need to drive the car to the car wash (because they hadn’t yet hill-climbed that particular brain-teaser).

The law learned to grapple with this long, long ago. For example, res ipsa loquitur (1863) seems apt.