Impact of SB 1047 on Academic AI Research

Academia faces significant challenges under SB 1047, limiting access to AI technology and hindering the training of future leaders.

### The Impact of SB 1047 on Academia and Open Source Innovation
* 📚 " Good policy should **solve for the harm it's trying to address**, yet SB 1047 fails to do so, posing a **significant threat to academic research and the open source community**.
* 🔍  The bill's restrictions hinder the ability of startups and major players like Mata to **analyze and understand AI models**, which is **essential for societal impact assessments**.
* 💔  Without access to AI models, the next generation of AI leaders may not receive the necessary training, making SB 1047 **counterproductive to educational goals**.
* 📊  Despite claims that SB 1047 targets only large players, its liability implications could **deter companies from releasing models**, cutting off **academic inquiry and innovation**.
* 💼  Well-resourced universities may wish to invest in AI research, but the **financial burden of compliance and legal fees** could stifle these efforts.
* 🖥️  The disparity in access to AI resources is evident, as academia struggles with **limited computing power** compared to larger entities like Mehta, which seeks **300,000 units**.
* 🏛️  Advocacy for the Create A.I. Act is crucial, as it would **subsidize computing and data for academic research**, which is **urgently needed for training future AI experts**.
* 🤝  The ongoing consultation regarding SB 1047 highlights widespread concern from academia and startups alike, as the bill may inadvertently **fortify established companies** while sidelining **smaller innovators**.

Many would say, well, that's a good thing, being able to control technology that some are fearful of. Why is this a threat to your academic research? Well, good morning, Caroline, and thank you for having me. I think what's really important about good policy is it should solve for the harm that it's trying to solve for. In the case of SB 1047. And unfortunately, it does not do that. However, what is most concerning from my perspective is it harms academia and harms academia because it hurts the open source community and makes it very difficult for the open source community startups or other industry or large players like Mata, who puts out the llama model for us to be able to dissect, understand how that works, what its impact will be for society. And we need to be able to train the next generation to be AI leaders. And if we can't be able to have access to this because this could hurt this

entire community, then I think SB 1047 is actually counterproductive. Now, what those who have written SB 1047 would come back with, and indeed we did have Scott Wiener on the show previously saying, look, we're not taxing, we're not asking those that are running smaller models who are spending less than 10 million or so. We are only after the big players. Why is it such a crucial impact, for example, when Llamas seems to be going great, we just had numbers out, a matter that there's 20 million times downloads in the past month alone. Right. And that's precisely the point. Mita alone right now is looking at maybe restricting llamas use in the EU because of the overly stringent areas. SB 1047 is by far more far-reaching than this when it comes to the liability issues related to that. And the problem with that is, is companies like MARTA won't be releasing

this if there are such strong liability that can come against them for this. And without that, we won't be able to understand and interrogate these models and see how they work, the way academia does and how we do that very well. So that will cut off an entire choke point to us. The second I would also add that academia itself, well-resourced universities, do want to look into how to build these. And so there's consortiums of universities that might want to spend more money in building these so we can train the next generation on how this works. And we won't be able to do that because we won't be able to afford all of the lawyers and the legal fees that will be required to be able to stay in compliance. Something that you just said, Russell, about sort of academic access to technology. You know, we had Nvidia's earnings last night in a discussion with Jensen Huang. I had was about the broadening out of

the use of AI accelerators beyond the Hyperscalers can Stanford Institute for Human-centered I get access to a cluster like is this a situation that works for you. It's it's actually a significant challenge for us. We actually just for able to secure 30/100 and that sounds wonderful, but it's unfortunate because the reality is is if you look at what Mehta has tried to secure, they're trying to secure 300,000. So the reality is, is academia is constrained as to what it can do in this particular space. And we are very limited. And what we need to do is have Congress pass the Create A.I. Act that is a government subsidizing compute and data for academic research, and it's desperately needed. And academia also needs to consider how to reform here and put a greater degree of investment in this so that we can be able to train students so that they can be the next day.

AI leaders. SB 1047. Sorry to go back to it. Is this the end of the story, or do you think you'll continue to be consulted as this moves forward? We are being continually consulted on this, but the reality is, is there is a chorus of people who are deeply concerned about this. It's not just academia. It is indeed small businesses that are startups that are afraid of what this will mean for them. And there's this ironic side of this. It bizarrely will entrench or make the most powerful companies even more secure in this space because they can afford the legal fees, they can afford all of the compliance fees, and things that come from this. Startups in academia cannot.

CA AI Safety Bill Advances Despite Critics Concerns

Welcome to the Bloomberg Technology YouTube channel. Subscribe for original series and mini-docs on game-changing technologies, profiles on innovators, live conversations with experts, and breaking technology news & analysis from Bloomberg TV.

Bloomberg Technology

The Impact of SB 1047 on Academia and Open Source

SB 1047 is criticized for harming academia and the open source community, making it difficult for startups and industry players to access and understand AI models.

The bill's focus on taxing only big players is seen as counterproductive, as it could restrict access to crucial resources for training the next generation of AI leaders.

The far-reaching liability issues of SB 1047 may lead to companies like MARTA restricting the use of models like Llama, further exacerbating the problem.

Strong liabilities against AI companies hinder the ability to interrogate and understand their models, cutting off a crucial point of access for academia.

Well-resourced universities are interested in building AI models but may struggle to afford the necessary legal compliance.

The high legal fees could impede the training of the next generation in AI technology.

Challenges in Securing AI Accelerators for Academia

The Stanford Institute for Human-centered AI is struggling to secure sufficient AI accelerator resources, with only 30 out of 100 needed.

In contrast, organizations like Mehta are trying to secure 300,000, highlighting the constraints faced by academia.

There is a urgent need for government support through the Create A.

Act and for greater investment from academia to train students for future roles in the field.

The consultation on SB 1047 is ongoing, with widespread concern from various sectors, including academia and small startups.

The legislation may inadvertently strengthen the position of powerful companies that can afford the associated legal and compliance costs.

In contrast, startups and academic institutions may struggle to cope with the financial burden, potentially exacerbating the inequality in the industry.