The money pit
OpenAI is reportedly making over $1.3 billion in annualized revenue, 30% more than previous estimates.
Why it matters:
While $1.3 billion is impressive, OpenAI is spending an absurd amount to keep ChatGPT running and to train new models, and likely isn’t profitable.
For months, people have been wondering whether AI companies are just another VC money pit - and so far, the only clear winners have been hardware and infrastructure companies.
GitHub Copilot, for example, was reported to be losing $20/month per user - but insiders now say it's generating $100 million in ARR, and is not losing money.
Elsewhere in OpenAI:
Techcrunch reports that OpenAI’s mobile revenue growth is starting to slow.
Next month is OpenAI's first developer conference, where it will supposedly cut API costs, release APIs for GPT-V, and launch a new memory storage product.
And according to FCC filings, Sam Altman is the largest investor in Humane, the secretive AI startup releasing its AI Pin next month.
Genie, meet bottle
The US Department of Commerce is considering a new blockade for general purpose AI programs - so-called “frontier models.”
Between the lines:
The plan could slow AI innovation in the US, as American AI researchers often collaborate with their Chinese counterparts.
Previously, the Commerce Department enacted export controls on sending chips to China. Now, the Biden administration is looking at closing a loophole that allows subsidiaries to purchase them.
And academics in Saudi Arabia worry that collaborations with China could jeopardize their access to US-made chips as the country rushes to build LLMs.
Elsewhere in AI geopolitics:
After the attacks in Israel, Intel, Nvidia, and other chipmakers are planning to boost security at their Tel Aviv locations.
China outlines its plan to get to 300 exaflops (up from 220 exaflops this year) of computing capacity by 2025.
And a leaked draft of an upcoming "guide to AI governance" from the Association of Southeast Asian Nations (ASEAN) shows a hands-off approach to AI regulation, in direct contrast to the EU.
People are worried about AI clones
A new bipartisan bill - the No Fakes Act - seeks to protect actors, musicians, and other performers from unauthorized digital replicas of their faces and voices.
Why it matters:
Many states have "likeness laws" to stop unauthorized use of someone's identity, but they vary widely and have no federal equivalent.
Of course, that hasn't stopped groups like the RIAA from trying to get voice cloning sites on government piracy watchlists.
Yet even when it comes to legal, licensed clones, it's still a brave new world. From SB Nation: " Facebook’s AI Tom Brady is a weird creep who’s obsessed with Travis Kelce".
Elsewhere in AI anxiety:
Adobe has a new icon to encourage tagging AI-generated content.
The US Space Force is the latest branch of the government to ban the use of LLMs due to security concerns.
And some are now raising concerns over the AI revolution's environmental impact.
Things happen
Character.AI launches Group Chat, for multi-AI conversations. AMD acquires Nod.ai, a model-deployment startup. LLMs can now detect and mimic empathy. Google's Search Generative Experience can now generate images. Man who attempted to assassinate the Queen was encouraged by a chatbot. Adobe debuts its Firefly Image 2 model, with big upgrades from V1. AI hype is built on flawed test scores. DeepMind's revenue fell 21% year-over-year. Google vows to defend users of its generative AI tools from copyright lawsuits. AGI is already here. Dropbox announces open beta for its AI-powered search tool. Microsoft's new tools for health care organizations, including an Azure AI Health Bot. The AI research job market.
Last week’s roundup
AI Roundup 035: Me or your lying eyes
Deepfakes have been around for a while, but advancements in generative AI have opened the floodgates to creating them at scale.