tomo,
Yeah, the pollies have a pretty limited repertoire. Something sensational happens and all they can do is write some new rules or call an inquiry. E.g. all the anti-terrorism laws supposedly to prevent something that was already profoundly illegal.

Rather than put strictures on AI, maybe the EU (and everyone else) should set AI to work through the mountains of laws and regulations to highlight contradictions and inconsistencies. Then they can put their heads together to make it all a bit less ridiculous.

Thinking a bit more on that notion of asking the chatbot to give the case for its previous answer, that could be rounded out by asking it for the case *against* as well. That makes the AI a (notionally) impartial advocate. Still has to convince a human who'll put his neck on the line for the decision.

Depends a lot on the training. I suspect the unvetted training material (whole internet) for ChatGPT would lead to logical inconsistencies in the two cases (e.g. plenty of stuff describing renewable energy as the cheapest, and plenty describing it as the dearest).

Mailman,
It's fair payment for how badly the old media have behaved. It continues too. ABC's coverage last night of the Kennedy grilling gave accusations (from politicians and his cousin), but none of his responses. I don't expect ABC needed to doctor it much; the international feed would have had their slant covered adequately.

The mid-air collision over the Potomac: well covered here (another tomo regular source, I think).

Robbo,

I also love how the "new" media will be seated in the front of the press conferences 😂😂

tomo,
I agree: there will be applications where "AI" is beneficial, and it has put computers into creative areas in the role of creator, not merely a tool. That's new.

What you say about NVidia's use of "AI" in chip design is what I was getting at with the "shape shifter". There's a lot more than the neural net core sailing under the glorious "AI" banner. That promotes a wide perception of all-powerful AI. There was a comment at Jo Nova's recently saying there was no human job that AI couldn't do. I didn't respond (too late, all threads die quickly at Jo Nova's), but I'd say one safe job is taking blame. I don't think we're quite ready for computers to take legal responsibility.

Optimistically, the EU AI Act might be an acknowledgement of this, to sober things up a bit, and to clamp down on outright fraud a la Theranos. Then you consider who's writing the new laws and who's signing off on them, and have to feel they'll just be waving the carpet-baggers through.

I was chatting with a friend last night. With the recent excitement, he (unlike me) had made some effort looking into Deepseek. One very interesting thing that is claimed is that it *can* explain the reasoning behind its answers. If true, that is a dramatic change. Given the full reasoning, a *human* is in a position to be convinced (or not), and to put himself in the position of decision-maker and take responsibility.

Given the one-way nature of the neural net matrix operations (AIUI), I suspect Deepseek's explanation is not really giving its "reasoning", but the answer to a *new* query: Give a rationale for *previous answer*, and the machine synthesises that rationale with no reference at all to how it came to the first answer. That's not altogether disturbing — humans quite often start with a hunch, then form a rationale post-hoc — but it's not going to be an exercise in cold logic.

At least that's my hunch :-)

Thanks for the asianometry tip. Another place to explore (how do you find them all, and find the time to watch them?!) Also, always enjoy a Matt cartoon.

Watched the Fake News Awards. Started out well, with things that were clearly false, and known to be false at the time of original broadcast. The later stuff was more complicated. The truth is a pretty slippery thing in the baddies vs baddies conflicts.

Mailman,
A surprising bit of news from the OMB bunker was that they had reinstated 440 (IIRC) press passes that had been revoked by Biden. I don't think we heard a peep about Biden's massive attack on the fourth estate. It's almost like the fourth estate hasn't been doing its job...

I see over at G Fawkes they are highlighting Labour forcing Cuadrilla to concrete its last viable shale gas wells. We are screwed unless Uncle Don's American gas and oil drives the prices down very soon!! Then lets see Labour spin how Orange Man Bad saved britain from going dark! 😂

The left really does have a hard on for control doesnt it.

Wont be long before any wrong thought will leave one branded as a right wing extremist...oh wait, Labours on that already!

tomo,
I suppose it may not have been exactly consensual... a bit like Mugabe stealing a handshake from Jack Straw a while back. Starmer's wins on the revolting scale though.

As for the AI thing, I think (hope) it's the start of the bubble bursting. Was not greatly impressed with that fellow's explanation of Deepseek. Metaphors of "master craftsmen training apprentices" work well rhetorically, but what does it come down to in practical terms? Seems like it amounted to training on a smaller, more focussed and accurate data set.

That certainly helps for a narrow problem. Thing is, AI is being pushed as a giant leap forward from search engines: ask a question, here's the answer. Scaling up Deepseek's approach, we just need to bring up the quality of the global training set -- i.e. the whole internet. Quite a big task I think. What's more, once done, a plain search engine will be enough to reveal truth.

One kind of creature that appears in the horror genre is the shape shifter: the monster that uses your imagination so it can appear safe and unthreatening. The AI bubble is built on the same sort of thing. People have heard about neural nets, but those are a bit confusing. They assume others have checked it all out and just go ahead and use their own notions of "intelligence".

There are already some nifty things that have come from it, but I'm very confident there won't be anything to justify the amount of money being spent. Fortunes will be lost.

There was a comment at Jo Nova's yesterday about using AI to develop new silicon wafers for transistors. *Some* of them performed better than human-designed ones and the researchers couldn't explain why. Yes. Neither could the AI, I'm sure. The magic answers machine is worse than useless when its answers are sometimes bad. Croesus consults the Oracle: If you go to war, a great empire will fall. Not going to say which empire though.

Mailman,
Was perplexed wondering why you wanted to defend the US Office of Management and Budget, then it dawned on me who OMB was.