AI digest: Big money, small models, and security slip-ups
OpenAI raises massive funding, Google cuts video costs, small models punch above their weight, and Anthropic has another bad week.
OpenAI’s funding rounds keep getting more ridiculous, Google’s making AI video affordable, and security incidents are becoming a weekly occurrence.
OpenAI raises $3B from retail investors in $122B round
OpenAI pulled in another $3 billion from retail investors as part of a monster $122 billion funding round led by Amazon, Nvidia, and SoftBank. The company’s now valued at $852 billion as it gears up for an IPO. These numbers are getting so big they’re hard to take seriously, but the retail investor angle suggests OpenAI’s trying to build public market appetite early.
Google’s Veo 3.1 Lite slashes video generation costs
Google released Veo 3.1 Lite, cutting AI video generation costs by more than half whilst maintaining speed. Finally, someone’s tackling the pricing bottleneck that’s kept generative video stuck in demo land. This could actually make AI video practical for production use, not just flashy marketing campaigns.
Liquid AI’s 350M parameter model challenges scaling orthodoxy
Liquid AI’s LFM2.5-350M packs serious performance into just 350 million parameters through massive training on 28 trillion tokens plus reinforcement learning. It’s a proper challenge to the “bigger is always better” mentality that’s dominated AI development. Smart training beats brute force parameter counts, apparently.
Anthropic has another terrible week
Anthropic accidentally published Claude Code’s source code for anyone to find, following last week’s internal blog post leak about their Mythos model. Two major security cock-ups in one week suggests they need to sort their internal processes rather urgently. Being the “safety-focused” AI company whilst leaking your own code is not a great look.
Mercor hit by cyberattack through LiteLLM compromise
The AI recruiting startup confirmed a security incident after hackers compromised the open-source LiteLLM project to access company systems. Supply chain attacks on AI tooling are going to become a massive headache as more companies build on these frameworks.