AI Roundup 060: Another CEO gone
AI Roundup 060: Another CEO gone
March 29, 2024.
Another CEO gone
Last Friday, Emad Mostaque, founder and CEO of Stability AI, resigned.
The big picture:
Mostaque's resignation comes after several other Stability AI departures (including key researchers behind Stable Diffusion), though tensions may have been simmering behind the scenes for months.
On Twitter, Mostaque has posted about pursuing "decentralized AI." Other reports suggest he may collaborate with his former head of strategic partnerships.
This makes the second high-profile AI CEO departure in as many weeks. The sky-high costs of training foundation models are causing plenty of turmoil - and potentially some industry consolidation.
Elsewhere in model mayhem:
Databricks released DBRX, an open LLM that surpasses Llama 2, Mixtral, and Grok on various benchmarks.
A121 Labs launched Jamba, an AI model that combines two architectures: transformers and state space models (SSMs).
And Claude 3 Opus surpassed GPT-4 on the LLM Chatbot Arena, the first time a model has done so since GPT-4 was released.
The AI talent war
As some AI CEOs resign, others ramp up the fight for talent. Tech companies are throwing around million-dollar annual compensation packages, making interviewing optional, and offering many other perks.
Why it matters:
This new wave of generative AI hype has created a shortage of those with hands-on experience researching, training, and optimizing LLMs. Mark Zuckerberg is going so far as to personally email AI researchers to try and recruit them.
It also means that researchers, like GPUs, will accrue to those with the biggest war chests - Big Tech companies. Smaller startups, as well as federal agencies, are going to have a harder time recruiting.
And the talent shortage could become a geopolitical issue - a new study has found that China has produced nearly 50% of the world's top AI researchers, compared to 18% from the US.
Elsewhere in AI anxiety:
Scammers are using sample videos of influencers to create unauthorized AI deepfake ads, often pushing offensive products and ideas.
A wave of free-to-use "mental health chatbots" is taking hold with Gen Z, though no one yet knows whether the AIs are effective.
And at least one Telegram user is offering to create deepfake porn of anyone, for as little as $10.
UXL vs Nvidia
Currently valued at $2.2 trillion, Nvidia is nearly synonymous with AI chips. Now, several competitors have formed the Unified Acceleration Foundation (UXL) to try to break Nvidia's dominance.
Between the lines:
The UXL is composed of Intel, Google, Arm, Qualcomm, Samsung, and other tech companies. Their plan is to take on CUDA, Nvidia's software platform that has locked in AI developers.
Yet even with Nvidia's dominance, there's still a massive GPU shortage. Plenty of upstarts, the latest being MatX, are fighting for their slice of the pie.
And as we continue to pour money into bigger and better chips, we could end up with a trillion transistor GPU.
Elsewhere in the FAANG free-for-all:
Google.org, the charitable arm of Google, debuted an accelerator for generative AI.
Azure AI Studio adds tools to help prevent users from "jailbreaking" AI chatbots.
Amazon spends another $2.75B on Anthropic, the second tranche of its planned $4B round.
Meta is adding AI features to its Ray-Ban glasses this April.
And Adobe unveiled new AI features at its annual Summit conference, including Firefly structure-based prompting, custom Firefly models and services, and GenStudio - a brand-safe AI generator for marketers.
Things happen
The BBC received complaints after using AI in its Doctor Who marketing. Teams is getting smarter Copilot AI features. How Silicon Valley’s ‘Oppenheimer’ found lucrative trade in AI weapons. A look at the international battles to regulate AI. Airtable launches new AI features. Why AI search engines really can’t kill Google. OpenAI's first Sora videos from filmmakers and artists. The FT tests an AI chatbot trained on its own articles. Moirai: a time series foundation model for universal forecasting. AI and data infrastructure drives demand for open source startups. Israel quietly rolled out a mass facial recognition program. AI robots slow the spread of Dutch tulip infections. The first Rabbit R1s are expected to arrive at the end of April. AI hallucinates software packages and devs download them. I scraped all of OpenAI's Community Forum. NYC will test AI gun detectors on the subway. Google adds AI-generated trip itineraries to search and Maps navigation to Gemini. A pink slime site used AI to rewrite our AI ethics article. The Air Force bought a surveillance-focused AI chatbot.
Last week's roundup
