Some laws are just bad. When the API-mediated/closed-weights companies agree with the open-weight/operator-aligned community that a law is bad, it’s probably got to be pretty awful. That said, though my mind might be playing tricks on me, I seem to recall the big labs being in favor at one time.
There are a number of related threads linked, but I’ll personally highlight Jeremy Howard’s open letter as IMHO the best-argued case against SB 1047.
> When the API-mediated/closed-weights companies agree with the open-weight/operator-aligned community that a law is bad, it’s probably got to be pretty awful.
I’d be careful with that cognitive bias, because obviously companies dumping poison into water sources are going to be opposed to laws that would prohibit them from dumping poison into water sources.
Always consider the broader narrative in addition to the specific narratives of the players involved. Personally, I’m on the side of the fence that’s grumpy Newsom vetoed it, because it stymies the larger discussion about regulations on AI in general (not just LLMs) in the classic trap of “any law that isn’t absolutely perfect and addresses all known and unknown problems is automatically bad” often used to kill desperately needed reforms or regulations, regardless of industry. Instead of being able to build on the momentum of passed legislation and improve on it elsewhere, we now have to deal with the giant cudgel from the industry and its supporters of “even CA vetoed it so why are you still fighting against it?”
As someone who was once asked under oath "What did you mean when you sent the email describing the meeting as a 'complete clusterfuck'?" I can attest to the wisdom of those words.
It’s probably a google search away, but if I’ve typed it slack/outlook/whatever, but not sent it because I then thought better of it, did the app still record it somewhere? I’m almost sure it has to be and I would like to apologize in advance to my senior leadership…
That depends greatly on your tooling, your company, as well as the skills and ethics of your Enterprise IT team.
Generally speaking, it’s in ours’ and the company’s best interests to keep as little data as possible for two big reasons: legal discovery and cost. Unless we’re explicitly required to retain historical records, it’s a legal and fiscal risk to keep excess data around.
That said, there are situations where your input is captured and stored regardless of whether it’s sent. As you said, whether it does or not is often a simple search away.
> The definition of “covered model” within the bill is extremely broad, potentially encompassing a wide range of open-source models that pose minimal risk.
Who are these wide range of >$100mm open source models he's thinking of? And who are the impacted small businesses that would be scared to train them (at a cost of >$100mm) without paying for legal counsel?
There were a lot of questionable Federal laws that made it through in the 90s, such as DOMA [1], PRWORA [2], IIRIRA [3], and perhaps the most maddening to me, DSHEA [4].
My understanding is that tech was politically weaker back then. Although there were some big tech companies, they didn’t have as much of a lobbying operation.
As I remember it, among other reasons, tech companies really wanted “multimedia” (at the time, that meant DVDs) to migrate to PCs (this was called the “living room PC”) and studios weren’t about to allow that without legal protection.
No snark, but what's wrong with the DMCA? From what I understand it, they took the idea that it's infeasible for a site to take full liability for user-generated copyright infringement (so they granted them safe harbor), but that they will be liable if they ignore take down notices.
"Thanks to fair use, you have a legal right to use copyrighted material without permission or payment. But thanks to Section 1201, you do not have the right to break any digital locks that might prevent you from engaging in that fair use. And this, in turn, has had a host of unintended consequences, such as impeding the right to repair."
The biggest problem with it, AFAICT, is that it allows anyone who claims to hold a copyright to maliciously take down material they don't like by filing a DMCA notice. Companies receiving these notices have to follow a process to reinstate material that was falsely claimed, so many times they don't bother. There's no mechanism to punish companies that abuse this.
The bill included language that required the creators of models to have various "safety" features that would severely restrict their development. It required audits and other regulatory hurdles to build the models at all.
If the danger is coming from the amount of compute invested, then cost of compute is irrelevant.
A much better objection to static FLOP thresholds is that as data quality and algorithms increase, you can do a lot more with fewer FLOPs / parameters.
But let’s be clear about these objections - they are saying that FLOP thresholds are going to miss some harms, not that they are too strict.
The rest is arguing about exactly where the FLOP thresholds should be. (And of course these limits can be revised as we learn more.)
Cost per FLOP continues to drop on an exponential trend (and what bit flops do we mean?). Leaving aside more effective training methodologies and how that muddies everything by allowing superior to GPT4 perf using less training flops, it also means one of the thresholds soon will not make sense.
With the other threshold, it creates a disincentive for models like llama-405B+, in effect enshrining an even wider gap between open and closed.
Why? Llama is not generated by some guy in a shed.
And even if it were, if said guy has such amount of compute, then it's time to use some of it to describe the model's safety profile.
If it makes sense for Meta to release models, it would have made sense even with the requirement. (After all the whole point of the proposed regulation is to get some better sense of those closed models.)
“But the big guys are struggling getting past 100KB, so ‘think of the small guys’ doesn’t make sense when the limit is 640KB.”
How do people on a computer technology forum ignore the 10,000x improvement in computers over 30 years due to advances in computer technology?
I could understand why politicians don’t get it.
I should think that computer systems companies would be up in arms over SB 1047 in the same way they would be if the government was thinking of putting a cap on hard drives bigger than 1 TB.
It puts a cap on flops. Isn’t the biggest company in the world in the business of selling flops?
It would be crazy if the bill had a built-in mechanism to regularly reassess both the cost and FLOP thresholds… which it does.
Inversely to your sarcastic “understanding” about politicians’ stupidity, I can’t understand how tech people seem incapable or unwilling to actually read the legislation they have such strong opinions about.
It's troubling that you are saying things about the bill which are false, and then speculating on the motives of someone just pointing out that what you are saying is false.
What's false is the idea that the limit is going to be a burden on small companies, because you can ignore the flop limit if you're spending less than a hundred million dollars. (Big companies, in contrast, can use a percent of their budget for compliance.)
Being able to ignore the flop limit makes basically everything else you've said irrelevant. But just to quickly go through: I don't want to argue about what a 'startup' is but they're not 'small guys'. Advanced tech can be compensated for, but also it doesn't change the fact that staying under $100 million keeps you excluded. Export controls have nothing to do with this discussion, and they involve a completely different kind of 'flop limit'.
We periodically raise flop limits in export control law. The intention is still to limit China and Iran.
Would any computer industry accept a government mandated limit on perf?
Should NVIDIA accept a limit on flops?
Should Pure accept a limit on TBs?
Should Samsung accept a limit on HBM bandwidth?
Should Arista accept a limit on link bandwidth?
I don’t think that there is enough awareness that scaling laws tie intelligence to these HW metrics. Enforcing a cap on intelligence is the same thing as a cap on these metrics.
Has this legislation really thought through the implications of capping technology metrics, especially in a state where most of the GDP is driven by these metrics?
Clearly I’m biased because I am working on advancing these metrics. I’m doing it because I believe in the power of computing technology to improve the world (smartphones, self driving, automating data entry, biotech, scientific discovery, space, security, defense, etc, etc) as it has done historically. I also believe in the spirit of inventors and entrepreneurs to contribute and be rewarded for these advancements.
I would like to understand the biases of the supporters of this bill beyond a power grab by early movers.
Export control flop limits are designed to limit the access of technology to US allies.
I think it would be informative if the group of people trying to limit access of AI technology to themselves was brought into the light.
Who are they? Why do they think the people of the US and of CA should grant that power to them?
Wait sorry, are you under the impression that regulated entities get to “accept” which regulations society imposes on them? Big if true!
Your delusions and lack of nuance shown in this very thread are exactly why people want to regulate this field.
If developers of nuclear technology were making similar arguments, I bet they’d have attracted even more aggressive regulatory attention. Justifiably, too, since people who speak this way can't possibly be trusted to de-risk their own behavior effectively.
Or used a model someone open sourced after spending $100M+ on its training?
Like if I’m a startup reliant on open-source models I realize I don’t need liability and extra safety precautions but I didn’t hear any guarantees that this wouldn't turn off Meta from releasing their models to me if my business was in California?
I never heard any clarifications from the Pro groups about that
You misread my intention, I think. I was pointing out that startups that don't raise enough to train a GPT-4 are excluded from the bill, that requires >$100mm expenses for a model to be covered.
If there was just one quasi-monopoly it would have probably supported the bill. As it is, the market leaders have the competition from each other to worry about. Getting rid of open models wouldn't let them raise their prices much.
So if it's not them, who is the hidden commercial interest sponsoring an attack on open source models that cost >$100mm to train? Or does Wiener just genuinely hate megabudget open source? Or is it an accidental attack, aimed at something else? At what?
Like I said, supporters included wary copyright holders and the bottom-market also-rans like Musk. If your model is barely holding up against Llama, what's the point of staying in.
The market in my argument looks like OpenAI ~ Anthropic > Google >>> Meta (~ or maybe >) Musk/Alibaba. The top 3 aren't worried about the down-market stuff. You're free to disagree of course.
It's an interesting thought that as AI advances, and becomes more capable of human destruction, programmers, bots and politicians will work together to create safety for a large quantity of humans
I would note that Facebook and Google were opposed to eg gdpr although it gave them a larger share of the pie.
When framed like that: why be opposed, it hurts your competition? The answer is something like: it shrinks the pie or reduces the growth rate, and that's bad (for them and others)
The economics of this bill aren't clear to me (how large of a fine would Google/Microsoft pay in expectation within the next ten years?), but they maybe also aren't clear to Google/Microsoft (and that alone could be a reason to oppose)
Many of the ai safety crowd were very supportive, and I would recommend reading Zvi's writing on it if you want their take
Yeah, I think the argument that "this just hurts open models" makes no sense given the supporters/detractors of this bill.
The thing that large companies care the most about in the legal realm is certainty. They're obviously going to be a big target of lawsuits regardless, so they want to know that legislation is clear as to the ways they can act - their biggest fear is that you get a good "emotional sob story" in front of a court with a sympathetic jury. It sounded like this legislation was so vague that it would attract a hoard of lawyers looking for a way they can argue these big companies didn't take "reasonable" care.
Sob stories are definitely not covered by the text of the bill. The "critical harm" clause (ctrl-f this comment section for a full quote) is all about nuclear weapons and massive hacks and explicitly excludes "just" someone dying or getting injured with very clear language.
