[AI]■ STORY TIMELINE

ANTHROPIC INTRODUCES 'MODEL SPEC MIDTRAINING' TECHNIQUE

Anthropic researchers have detailed a new training approach called "model spec midtraining" that inserts an intermediate stage between pretraining and fine-tuning. The technique aims to improve how AI models generalize from alignment training.

10 SOURCESFIRST SEEN MAY 7, 02:20 AM► READ THE ARTICLE

Techmeme+0m

Anthropic researchers detail "model spec midtraining", which adds a stage between pretraining and fine-tuning to improve generalization from alignment training (Anthropic)

Anthropic: Anthropic researchers detail “model spec midtraining”, which adds a stage between pretraining and fine-tuning…

Techmeme+8h 6m

Elon Musk says he spent time with Anthropic "to understand what they do to ensure Claude is good for humanity"; Musk called Anthropic "evil" and "misanthropic" (Lauren Goode/Wired)

Lauren Goode / Wired: Elon Musk says he spent time with Anthropic “to understand what they do to ensure Claude is good f…

Techmeme+8h 15m

At Anthropic's developer conference, Dario Amodei says the company could grow by 80x in 2026, and its growth rate is "crazy" and "too hard to handle" (Natallie Rocha/New York Times)

Natallie Rocha / New York Times: At Anthropic's developer conference, Dario Amodei says the company could grow by 80x in…

Techmeme+9h 6m

Anthropic's SpaceX deal helps it address a severe compute deficit and comes amid Elon Musk's OpenAI lawsuit; Grok never grew to utilize xAI's Colossus 1 (Axios)

Axios: Anthropic's SpaceX deal helps it address a severe compute deficit and comes amid Elon Musk's OpenAI lawsuit; Grok…

Techmeme+16h 30m

While Anthropic will use the Colossus 1 data center, which has a really bad environmental record, xAI retains the larger Colossus 2 for its own AI training (Simon Willison/Simon Willison's Weblog)

Simon Willison / Simon Willison's Weblog: While Anthropic will use the Colossus 1 data center, which has a really bad en…

Techmeme+17h 41m

Anthropic researchers detail natural language autoencoders, which convert LLM activations, the numbers encoding a model's thoughts, into natural language text (Anthropic)

Anthropic: Anthropic researchers detail natural language autoencoders, which convert LLM activations, the numbers encodi…

Techmeme+22h 30m

Mozilla says Anthropic's Mythos Preview and other AI models helped it identify and ship 423 Firefox security bug fixes in April, compared to 31 a year earlier (Russell Brandom/TechCrunch)

Russell Brandom / TechCrunch: Mozilla says Anthropic's Mythos Preview and other AI models helped it identify and ship 42…

Techmeme+1d 2h

Sources: Anthropic weighs raising as much as $50B at a ~$900B pre-money valuation, likely closing within two months; its annualized revenue is nearing $45B (Financial Times)

Financial Times: Sources: Anthropic weighs raising as much as $50B at a ~$900B pre-money valuation, likely closing withi…

Techmeme+1d 16h

Akamai says it struck a seven-year cloud computing deal with a "leading frontier model provider"; sources: the deal was with Anthropic and is worth $1.8B (Rachel Metz/Bloomberg)

Rachel Metz / Bloomberg: Akamai says it struck a seven-year cloud computing deal with a “leading frontier model provider…

Techmeme+2d 3h

Anthropic details how it improved Claude's safety training after finding agentic misalignment in older models, such as Opus 4 blackmailing engineers (Anthropic)

Anthropic: Anthropic details how it improved Claude's safety training after finding agentic misalignment in older models…

◄ BACK TO ARTICLE