New meta-analysis (2024) finds human-AI collaboration is task-dependent: performance increases for creative & open-ended tasks (like content creation), but performance drops for decision-making tasks; human-AI collaboration seems to work best when humans already outperform AI
I'm now of the opinion that LLM use should be fairly limited for most employees.
The amount damage caused by uncritical acceptance of hallucinations is far greater than I'd have expected. (and I was thinking it'd be pretty bad... it's worse.)
I'm increasingly concerned with signals from @EU_Commission that Digital Services Act is a tool for censorship. It was never introduced as such during works. Now it's passed and some EU policymakers appear to intend to use it as such. Unfortunate #DigitalEUambassador
GenAI in education: essay generated from the plagiarising machine➡️passed through AI to check plagiarism➡️“read”, “reviewed”, & “graded” by AI
the ONLY party that benefits from all this is big tech/AI vendors while education is destroyed & the next generation becomes dumber
Our Canadian friends may appreciate this 2021 document analyzing Google's programmatic business north of the border including the critical trojan horse of GA4 to enable user consent if privacy laws advance. /27 https://t.co/Udg1T93bPq
Google sees hard paywalls for content (like @ft) as a threat to their goals because they can’t scrape the content for their own profit.
They also see all the privacy tools we use (ad blockers, private browsing) as a direct threat to their business.
We knew; just confirmation.
wow. an amazing 325 page google strategy document quietly unsealed buried in google antitrust docket. It's gonna take a long thread but I have pulled out the gems. It's from 2017 planning, no doubt Google will just say these were only ideas but many will look very familiar. /1
Four women in my beautiful country Iran are at risk of execution for the crime of wanting freedom, democracy, equality, and dignity.
Do you expect me to mourn the killing of a terrorist instead? 😡
Outrage should be for women being killed ny those
terrorists simply for showing their hair!
Had an exciting day seeing the @WhiteHouse from the inside to talk about sociotechnical AI safety research! A star-studded visit thanks to a chance view of the VP. Some of the relevant work: https://t.co/TOTqh6HDPH
Feedback to @Google.
Please don't waste electricity on this Al-constructed BS. I just wanted a simple search, in fact…the spelling.
I say "please" but it is a crime against the planet… please though any human reading this do everything you can to make this optional.
#AIEthics
Posting a video that shows Harris saying politically damaging things she never said, without making it very clear it's a parody, seems like the definition of defamation.
So here we are. Deepfakes being used to manipulate an election.
I guess no one is surprised, but the fact it's spread not by bots or anons but the very owner of @X should probably ring alarm bells.
I'm delighted to be able to announce that my next book At Dawn They Came: Soviet Terror and Repression 1917-1953 is scheduled for publication by @HofTotalitarian at the end of September.
It recounts the history of Soviet repression from Lenin's first labour camps through to their fully fledged climax as the Gulag system under Stalin. 🧵
(Yes, the labour camp system began with Lenin, not Stalin. Lenin is sometimes portrayed as the 'good communist' whose ideals were betrayed by Stalin. This, as my book shows, is nonsense. )
After a lot of argument, Claude is willing to teach me how to filet a fish by creating interactive tools, but only once I certify that I have handled the fish in an ethical way. And even then it is only willing to go so far.
As models get smarter, guardrails become more complex.
I love how Microsoft have made this so much about Crowdstrike. No amount of trying to undermine EU AI regulation is going to get the US tech sector out of having to act like mature infrastructure suppliers, including being responsible for their supply chain.