AI Roundup 080: Grok 2
August 16, 2024.
Workshops
There’s two weeks left to RSVP to the upcoming Midjourney Masterclass! If you’ve only ever seen Midjourney used via Discord, there’s a ton of new features to catch up on!
Grok 2
This week, xAI released two new models, Grok-2 and Grok-2 mini. Currently in beta, they’re available to paid X subscribers.
Between the lines:
The blog post claims that Grok 2 outperforms Claude and GPT-4 on the LMSYS leaderboard, but there are reasons to be skeptical of such claims.
Grok 2 is making waves, but for less-than-ideal reasons - users are generating offensive and violent images with few guardrails.
But Grok will not use training data from European users, as Elon Musk faces legal action for funneling external resources to xAI.
Elsewhere in frontier models:
OpenAI has reportedly released a new GPT-4o model in ChatGPT, with some users claiming improved step-by-step reasoning capabilities.
An overview of the Gemma model family architectures, explaining their design and capabilities.
Claude now supports prompt caching to reduce the costs and latency of large inputs.
And Lambda Labs unveiled Hermes 3, the first fine-tuned version of Llama 3.1 405B.
Fake news
In recent weeks, Kamala Harris has triggered newfound excitement (and turnout) in Democrats. Perhaps unsurprisingly, Donald Trump has falsely claimed Harris was faking her crowd sizes with AI.
Why it matters:
Fears over AI disinformation have loomed large this year, with billions of people preparing to vote in worldwide elections.
With a couple of exceptions, AI deep fakes haven't impacted the US Presidential election, but companies like Microsoft continue to find potential threats from foreign actors.
And if AI ultimately degrades trust in US institutions, it could have a knock-on effect on the rest of the world.
Elsewhere in AI anxiety:
A US judge ruled that artists can pursue some copyright infringement claims against Stability AI and Midjourney in their lawsuit.
The US FTC is banning the sale of fake reviews and testimonials, including those generated by AI.
MIT has released a comprehensive database of over 700 unique AI risks.
And new research shows how attackers can manipulate Copilot to automate spam and fraud.
Pixel 9
Google's Pixel 9 event was this week; while it was ostensibly about the company's new hardware, you can bet there was a ton of AI news, too.
All the announcements I could find:
Gemini will replace Google Assistant as the default on the new Pixel 9 lineup.
Gemini Live, Google’s answer to ChatGPT's Voice Mode.
Pixel Screenshot lets you save, organize, and query the screenshots on your phone.
Pixel Studio, which offers text-to-image features and AI editing.
AI-powered camera capabilities, like "Add Me", let users edit themselves into group shots.
New Android accessibility updates, including AI-powered features for Magnifier and new Live Caption languages.
And a slew of other AI odds and ends for the Pixel 9 phone, Pixel Watch 3, and Pixel Buds Pro 2.
Elsewhere in Google:
Google makes Imagen 3 available to all US users through its ImageFX platform.
Some publishers say they can't block the Googlebot crawler for AI Overviews without hampering their discoverability in Search.
And Eric Schmidt, after blaming working from home for Google’s recent struggles, walked back his comments.
Things happen
University of California doctors use brain implants and AI to help an ALS patient speak again. SoftBank courted Intel for an AI chip to rival Nvidia, but talks fell through. Inside the Snowflake-Databricks rivalry and their shared fear of Microsoft. Foxconn to start shipping Nvidia's next-gen GB200 chips. Foxconn's Q2 profits surge on strong AI server demand. SAG-AFTRA strikes a deal with Narrativ for AI voice replicas in digital ads. AI-capable PCs made up 14% of all PC shipments in Q2, with Apple leading the pack. Chinese tech giants test Huawei's Nvidia H100 challenger. Chipmakers face a labor crisis despite massive investments. An AI system for automated scientific discovery. There's just one problem: AI isn't intelligent. Google's former CEO on why the company was blindsided by OpenAI. About half of job seekers are using AI chatbots to assist with applications. AI project Fragmentarium pieces together the Epic of Gilgamesh. Replika CEO Eugenia Kuyda on AI companions and the future of chatbots. Polymarket partners with Perplexity and Tako for AI-enhanced betting experience. LinkedIn users lament the rise of hustle-culture influencers and AI-generated content. Wikimedia Foundation execs discuss AI's impact on Wikipedia. Microsoft diversifies AI investments post-OpenAI board dispute. Cisco plans second round of layoffs in 2024, affecting thousands. How I won $2,750 using JavaScript, AI, and WD-40. Supabase introduces in-browser Postgres with an AI interface. Study shows slapping "AI" on products may drive customers away. Replika's CEO says it's fine if we marry AI chatbots.
