Eric Daimler

120+ lawyers sanctioned for AI hallucinations. Zero AI systems sanctioned for anything. The future is formal. (4/4)

When the system can't suffer consequences, accountability doesn't disappear. It transfers. To the operator. To the deployer. To the general counsel who signed off. (3/4)

A surgeon who removes the wrong kidney goes to prison. An AI that generates a false positive in a targeting decision gets a version update. The AI can't die. It can't be sanctioned. It can't lose a license it never held. (1/4)

A lawyer who fabricates a citation gets sanctioned. Suspended. Potentially disbarred. An AI that fabricates a citation gets... updated. Over 120 lawyers have now been sanctioned in American courts for submitting AI-generated briefs containing hallucinated case law. Fabricated judges. Invented statutes. Quotes attributed to real people who never said them. In every case, the pattern is identical. The AI produced output that looked authoritative. The lawyer trusted it. The court discovered the fabrication after the fact. This is not a technology failure. It is a verification failure. The legal profession has centuries of standards requiring that every citation be traceable to a real source. The AI systems generating those citations have no equivalent standard. None. And here is what should alarm every executive deploying AI in a regulated environment: the legal profession is one of the few domains that actually caught the problem. It has adversarial proceedings. Opposing counsel checks your work. Judges verify citations. Most industries don't have that. Your financial models, your compliance reports, your procurement analyses, your medical summaries. Nobody is checking the AI's homework. Not because they don't care. Because there is no infrastructure to check it against. Scaling up a dictionary won't teach it calculus. And scaling up your test suite won't catch a hallucination that looks exactly like the truth. The future is formal.

This multiplication was off by 120 million. Now imagine this is your financial report to the SEC. Your data migration. Your construction specifications. Not to the board: ‘close enough’ is not a compliance standard.

We Live Like Royalty and Don’t Know It. ("The electric grid, the public-water supply, the food-distribution network, and the public-health system took the collective labor of thousands of people over many decades. They are the cathedrals of our secular era. They are high among the great accomplishments of our civilization. But they don’t inspire bestselling novels or blockbuster films. No poets celebrate the sewage treatment plants that prevent them from dying of dysentery." thenewatlantis.com/publications/w…

The future is formal. (6/6)

The jobs of the future belong to those who think clearly enough to work with AI, not hope it will think for them. (5/x)

Can my kids think precisely enough to survive in an AI-first world? (1/x)

Our new rule: We stop implicitly asking customers to "imagine." Now we just show them. (5/5)

Before: An LLM-generated legal brief with fake citations. After: The same brief, provably verified by our neurosymbolic AI. The difference is a revelation (to us academic-trained ignorami) (4/x)

OMG. We have been marketing our neurosymbolic AI completely wrong. We were asking people to imagine the value. When we were academics, that is a high-status game. In business, it is stupid. (1/x)

Coded out this optical effect I'd seen last week... you should be able to pause the video any time, and the cube will disappear! Which kinda makes sense, as its not flipping the pixels, so you don't see the movement, but still its a cool one!

Thanks for good conversation @iuea_uganda instagram.com/p/DBeBCuJsybE/…