ICYMI: Top Stories from the Week

VETO: California Governor Gavin Newsom Strikes Down “AI Safety Bill” SB1047

SB 1047 was essentially the first real attempt from state legislators in the US to establish precedent in regulating AI. The bill had some notable proponents, including Elon Musk and Anthropic, but faced widespread criticism throughout Silicon Valley (and even Nancy Pelosi).

• Included in the proposal were some strict regulations for large-scale AI models — including mandatory shutdown capabilities (a “human button”), third-party audits, and significant fines for violations.

• Newsom rejected the bill, citing its overly broad scope, lack of risk-based approach, and potential to hinder innovation…especially for smaller AI models.

• The veto could signal a shift towards developing more nuanced, evidence-based AI regulations that consider the varying risk levels of different AI applications.

Open AI’s Dev Day Brings 4 Major Announcements

OpenAI unveiled a suite of interesting API features which generally appear aimed at lowering barriers to entry for developers working with their models.

• Realtime API: gives developers speech-to-speech capabilities, offering six distinct voices.

• Model Distillation: simplifies the fine-tuning process for smaller models.

• Prompt Caching: slashes costs by half and boosts response times by up to 80%, optimizing API efficiency.

• Vision Fine-Tuning: elevates image recognition capabilities through combined image and text training.

Google Is Officially Adding Advertisements To Their AI Search Results

Google has launched ads within its AI-generated search summaries, initially for mobile users in the United States. The ads appear alongside and within AI Overviews for queries with commercial intent, showcasing sponsored product suggestions.

• Ads are displayed under a 'sponsored' label within AI-generated summaries

• The rollout is limited to mobile users in the U.S. for queries with commercial angles

• Google is adding inline links to sources in AI summaries, potentially increasing traffic to websites

• The company will not share ad revenue with publishers cited in AI Overviews

Open AI’s Canvas Release Turning Heads

Canvas is a new interface that opens in a separate window alongside the traditional ChatGPT chat window, providing a dedicated workspace for users to collaborate with the AI on more complex writing and coding tasks.

What’s new:

• Real-time editing, AI suggestions, formatting tools, code generation/review, version control.

• Improved contextual understanding with section highlighting for focused assistance.

• Canvas streamlines complex tasks requiring multiple revisions.

• Beta release ongoing for ChatGPT Plus and Team users; wider rollout planned soon for premium, eventually heading to all free users.

SoftBank Invests $500m into OpenAI

As part of a $6.6 billion funding round valuing the company at $157 billion, the SoftBank investment signifies another major vote of confidence in OpenAI's potential in being the leader of this new industry.

• The investment connects OpenAI with SoftBank's extensive network, potentially opening doors to new collaborations and markets, especially in Asia.

• The funding should help accelerate the development of more sophisticated AI models and expand AI applications across various industries.

• Challenges remain, including OpenAI's projected $5 billion loss this year and potential governance issues as the company considers transitioning to a for-profit structure.

Other Headlines

• OpenAI raises $6.6 billion in fresh funding as it moves away from its nonprofit roots

• Google Antitrust Case Comes Too Late for Publishers

• Microsoft Copilot Voice is more human-like than ChatGPT — and it's free to all users

• NVIDIA Corporation (NVDA) CEO Jensen Huang Confirms Full Production of Blackwell AI Chip Amid “Insane” Demand

• AI bots now beat 100% of those traffic-image CAPTCHAs

❶

Why is everyone so hyped about NotebookLM right now? 🤨 🤔

NotebookLM was first launched over a year ago by Google as an AI study tool targeting students, but has picked up significant notoriety this past week for it’s new feature that let’s users make podcast style audio recordings from documents and other sources.

• Automated Study Aids: generates comprehensive study guides, summaries, and key topics from uploaded documents.

• Audio and Video Integration: makes summaries of YouTube videos and audio files.

• Sharable Audio Discussions: The new Audio Overview feature allows users to turn their documents into engaging audio discussions that can be shared via public URL.

While initially popular among educators and students, NotebookLM is now attracting significant interest from business professionals. Wired Magazine even did a full demo article going into detail on how to do it for yourself.

Is this really a “ChatGPT moment” for Google?

*Meh* — maybe we’re so totally desensitized now to cool new AI products dropping, after seeing them rain down on us from every direction amidst announcements of billions of dollars of funding pouring in from investors basically every week for the last year, that even a new feature with arguably as much intrigue as ChatGPT seems just, ok…?

If fortunes were reversed, and it was ChatGPT launching now, a year or so after NotebookLM’s podcast generator kicked off the GenAI the market, we’d probably be saying the same thing about it instead.

I think the undercurrent here is that nothing seems unique at this point in the AI hype cycle. Questions of quality aside, this doesn’t feel like anything that the other models couldn’t do with just a little fine-tuning. It’s weird that it seems to have taken off more from user discovery than a trumpeted company-sponsored promotion, but I remain suspicious of the underlying substance to this hype.

Don’t get me wrong, the banter it creates between the AI voices is pretty cool. And I totally get that this can be applied beyond the novelty aspects for scenarios like creating a pseudo conversational form of learning (many of us seem learn better through dialogue than we do from just reading information) or condensing longer podcasts into shorter ones.

The use-case feels narrow to me, and I can’t get past it. I give NotebookLM podcasts a 1 month life expectancy. Somebody might master it, but the podcast space is so saturated already that I doubt we’ll see any AI chit chat coming for Joe Rogan’s crown any time soon.

~JL

❷

Should we care about OpenAI 💰💸becoming a for profit business?

As of October 2024, OpenAI has not yet fully transitioned to a for-profit model. However, the company is seriously considering this move and is in the process of restructuring. The exact timing of this shift has not been determined, as the board is still in the process of making a final decision.

OpenAI is considering converting its core structure to a public benefit corporation. This type of entity is designed to generate profits while also serving a social or public interest. The change would move OpenAI away from its current structure, which includes a non-profit board of directors overseeing a for-profit subsidiary.

What’s really going to change:

• Funding: changing to a for-profit entity will greatly increase OpenAI's ability to raise capital.

• Commercialization: increased focus on developing monetizable AI technologies.

• Governance: more control shifts to investors, away from the non-profit board's oversight.

• Regulation: a for-profit model may intensify regulatory scrutiny, particularly as OpenAI's influence in the AI industry grows.

Will AGI arrive faster?

Well, we must have seen this coming. With the pace that OpenAI set for developing AI, and the competitive landscape that sprung up around it, the bottomline was always zooming up behind them.

They simply need more money than they can realistically get as a non-profit, given their insane expenditures. The majority of their competitors are deep-pocketed enterprises, with plenty of resources, established cash flow, and a whole lot to lose if they end up last to the party with AI. Shifting to for-profit makes returns a lot more tangible for investors. And with the level of money they are throwing at OpenAI — returns have to be expected.

Now, there are plenty of us more conspiracy-minded folks out there who are now worried that OpenAI’s core mission of developing AGI for the good of humanity is going to be compromised…becoming “developing AGI for the good of the investors (at the expense of the rest of us)”. Elon Musk’s repeated lawsuits against OpenAI on those grounds certainly inflame that point of view.

But here’s an alternate take that might help us get some sleep in the last few years that we have to enjoy before terminators are permanently messing with our REM cycles…

I think shifting to a for profit model actually makes the achievement of AGI less likely, or at least shifts it to a future generation of humans that figure out how to truly value progress over profits.

The assumption is that AGI will be achieved faster because of all the money flooding into OpenAI. But that money comes with strings attached. I can’t truly claim to know the mindset of investors with enough cash to qualify for investing in OpenAI right now, but I don’t think its a stretch to assume that people dropping millions into OpenAI expect to see that money come back to them tenfold within their lifetime. That means OpenAI’s priorities have to shift to generating money, which is not necessarily a synergistic goal for achieving AGI.

Conventional wisdom would assume that companies driven by profit would develop the best possible products at the fastest possible rate to ensure that they continue being able to sell their products in a competitive landscape. But look at Apple and Samsung. Each year they roll out phones with a few minimal changes (like moving the camera to the other side of the phone, or letting us make weird avatars of ourselves) and they sell units because the driving force of consumption in today’s world isn’t seeking quality, it’s seeking what’s new and interesting. So companies just drip out changes and prolong development by years, maybe decades, to maximize the profits they can capture at each phase.

Apple might very well realize that the optimal form of the mobile computing and communications tool they helped pioneer is not actually the smartphone, but something else entirely like a headset. But the iPhone is so deeply ingrained in their revenue streams that they would be incredibly foolish to build something better than it before the market is finished consuming it.

It’s not a perfect correlation, but I think the same underlying logic holds for OpenAI and their pursuit of AGI.

~ JL

❸

What’s really behind Newsom’s veto ⚖️ of the AI Safety Bill?

Instead of doing a write up, how about I just drop a “deep dive podcast” from the famous NotebookLM instead? 😆

This is the article I used to generate the podcast.

I literally just pasted the link in their tool and clicked generate. Took about 3 minutes. See for yourself how it works!

~ JL