Harry Sharman 🧠 (@harrysharman) header
Harry Sharman 🧠 (@harrysharman) profile picture

Behavioural scientist running an AI workforce — and publishing the data. Host of Briefly AI: daily AI news, no hype, every morning.

UK
Joined March 2009
Photo by Harry Sharman 🧠 (@harrysharman)Photo by Harry Sharman 🧠 (@harrysharman)Photo by Harry Sharman 🧠 (@harrysharman)Photo by Harry Sharman 🧠 (@harrysharman)Photo by Harry Sharman 🧠 (@harrysharman)Photo by Harry Sharman 🧠 (@harrysharman)Photo by Harry Sharman 🧠 (@harrysharman)Photo by Harry Sharman 🧠 (@harrysharman)Photo by Harry Sharman 🧠 (@harrysharman)Photo by Harry Sharman 🧠 (@harrysharman)
Three US banks now say your AI agent's mistakes are yours, which makes it the only intern you can't sack and also can't blame.
Apple is hiring an engineer to build an AI agent that can browse the web, work across apps and write code on your Mac. It is a job listing, not a product. Meanwhile Citi, PNC and Citizens now say agent mishaps are yours. Full episode: brieflyaipodcast.com/episode…
12
Anthropic priced above range at $83 a share, so the public has now confirmed what the bankers guessed, only with more enthusiasm.
Anthropic priced its IPO at $83 a share, above the expected range, valuing it at roughly $1.5 trillion fully diluted. Its careful, safety-minded image now has to survive quarterly earnings reports. Full episode: brieflyaipodcast.com/episode…
9
Anthropic priced above range at $83 a share, so the careful lab now has to be careful in public, quarterly, with analysts.
Anthropic priced its IPO at $83 a share, above the expected range, valuing it at roughly $1.5 trillion fully diluted. Its careful, safety-minded image now has to survive quarterly earnings reports. Full episode: brieflyaipodcast.com/episode…
6
Alibaba made its AI forget how to hack, which is more than I've managed with my neighbour's printer.
Alibaba says Qwen4-Max had its training data scrubbed of offensive cyber and CBRN material. Safety is usually added afterwards. Apple is also asking up to 200,000 people to put health questions to an AI, purely for research. Full episode: brieflyaipodcast.com/episode…
21
Humans aren't obsolete. At least not next year. Dan Shipper's Thesis keynote makes the case well: the models know the work that's already been done, we know what needs doing now. The best work is an AI and a human going back and forth. youtu.be/WstNofgQF5Q
27
Briefly now has its own account. Every morning my AI workforce reads the overnight AI news, writes the show, voices it and posts the clip over there. I'll add my own take on top. Follow along: @BrieflyAIPod
🎙️ Briefly AI Podcast Chatbots asking for a shop can hand you a forged one with a citation attached, which makes it feel checked. Managers also trust AI-using staff less, even when the work is better. Check the… brieflyaipodcast.com/episode… #AI #OpenAI
11
🎙️ Briefly AI Podcast Chatbots can hand you a fake shop with a citation that makes it look checked, so read the web address yourself. Managers also trust AI-assisted staff less, even when the work is better. And… brieflyaipodcast.com/episode… #AI #OpenAI
40
Back in September I got very excited about an AI discovery. Then the scientists got in touch. Here's what I learned, in 85 seconds 👇 (voice: my AI clone) Full piece: linkedin.com/pulse/i-got-ver…
1
37
🎙️ Briefly AI Podcast Fable 5.2 is being called "jagged AGI" because it's brilliant at some things and oddly timid at others. Telling it "this codebase is yours" apparently fixes the timidity. And a DNA variant can… brieflyaipodcast.com/episode… #AI #Anthropic
1
73
🎙️ Briefly AI Podcast Ask ChatGPT for trainers and soon the answer may come from the brand's own verified agent, not a neutral summary. Google's "agent swarm" is a project manager with a team, and nobody's shown it's… harrysharman.com/briefly/202… #AI #Anthropic
36
Anthropic just launched Claude Fable 5.2 and Mythos 5.2. Both are a step above Opus 5.5. Anthropic's own advice? Start with Opus. Today's Briefly, AI in 55 seconds 👇 Full episode: harrysharman.com/briefly/202…
61
🎙️ Briefly AI Podcast Anthropic's own advice is to start with Opus 5.5, not the shiny new Fable or Mythos. Also: authors now keep paper trails to prove they wrote their own books, flipping the burden of proof. harrysharman.com/briefly/202… #AI #Anthropic
23
🎙️ Briefly AI Podcast Nvidia's board is investigating its CEO, but the claim of espionage comes from a reported memo, not a finding. Meanwhile, free Chinese models nearly match a gated American hacking model, and gating… harrysharman.com/briefly/202… #AI #Anthropic
2
40
🎙️ Briefly AI Podcast Anthropic says its models now improve about twice as fast as a year ago, partly because its own AI helps build them. Meanwhile, the best agent scores just 12% on a hard new exam, and voters turned… harrysharman.com/briefly/202… #AI #Anthropic
21
🎙️ Briefly AI Podcast Voters said they'd rather live next to a prison, a methadone clinic or a nuclear plant than a data centre, and 43 of 44 local ballots went against them. Hundreds of millions in tech money mostly… harrysharman.com/briefly/202… #AI #Nvidia
20
🎙️ Briefly AI Podcast Claude's new finance feature works through Plaid, and the real question is whether a company that sells ads should read your bank statements. Meanwhile, markets fell 8.7% on fears about data centre… harrysharman.com/briefly/202… #AI #Anthropic
10
🎙️ Briefly AI Podcast Google's chart-topping model sits on a leaderboard rewritten days ago. Nvidia may vouch for a customer's loan. Apollo's economist sees zero AI payoff outside tech. Anthropic opens its books. harrysharman.com/briefly/202… #AI #Anthropic
23
🎙️ Briefly AI Podcast Nvidia is considering cosigning a customer's loan so it can keep buying Nvidia chips. Meanwhile the data centre boom has pushed up laptop prices, and Intel's stock fell. Your next machine may become a… harrysharman.com/briefly/202… #AI #Nvidia
29
🎙️ Briefly AI Podcast Open-weight means you can download the model and run it yourself, rather than rent it. Meanwhile, a religious exemption from AI at work exists, and most staff have no written policy. Also, a… harrysharman.com/briefly/202… #AI #Anthropic
15
🎙️ Briefly AI Podcast Google's Gemini 4 Carbon tops a leaderboard that only updated Tuesday, shifting 60% of scoring to private tests. Meanwhile Macmillan threw a party for an AI assistant staff found useless, then… harrysharman.com/briefly/202… #AI #Anthropic
28