When all you have is a hammer, everything looks like a nail.

But did he really get asked about the issue of “not knowing”? He just saw it as a problem they could plug, no?

(To be clear: the general problem of “not knowing” and all the bad that comes out of it should of course be challenged much more often.)

That was my point: It's as far from a general solution to the general problem of “unknowns” as it gets. They're basically making it a “known” instead.

Also note that Altmann does *not* imply they would fix this particular timing issue by retraining the model to be “consciously ignorant”. They would much prefer to just add some kind of internal trick (access to a time API) to get time issue right.

(continued)
... but training a model on “unknowns” requires being able to list those unknowns first. That sounds like a completely daunting task, because you're essentially asking: “What is not in our training data?”

Isn't there a “first principles” problem here? An LLM basically works towards the “most likely outcome”, with some variability baked in. But even prompts with no “real” answer have a “most likely outcome”. The only thing one could do is train the model to answer “don't know” on very specific things.

This is... something. But also something that's going to happen a lot more often. 🍿

What is especially damning is how this Matt character is using ChatGPT as an “expert source” to prove something, which is simply journalistic/academic/intellectual malpractice. (Yet, it happens like... A LOT.)

Ook Vlaamse industrie zou wel varen bij ambitieus klimaatbeleid – MO* De klimaatdoelen nopen bedrijven tot gigantische investeringen, maar die komen nauwelijks van de grond.

De vraag die beleidsmakers beter aan de industrie zouden stellen: wie blijft er hier produceren bij toekomstige gasprijzen die structureel tussen de 30 en 60€/MWh liggen. Luister naar en richt ondersteuning op degene die daar "ja, wij" op antwoorden.
www.mo.be/analyse/ook-...

Pondcast #3: Calculate CO2 emissions with Thomas Goorden

♻️ Recycling plastic doesn't help the climate. At all. That's nor an opinion nor an Aprils fools joke. Recycling just delays incineration by a couple of years. The carbon ends up as CO2 either way.

Learn more in the latest Pondcast with @goorden.be

youtu.be/qE9GMVyRoEI

For those skeptical about this statement, here's the relevant homework, using the techniques discussed in the podcast:

apache.be/sites/defaul...

OK, “Dr. Heinroch Pocklock Krockpock”?

Checks calendar... right.

Your EarPeeps are broken!

What this guy says. Think a clanker can do what I do? Go ahead, try! I'll get the popcorn.

Stop Making Unit Mistakes | Data Frosch How attaching units to your calculations prevents costly mistakes — a lesson from the Mars Climate Orbiter and investigative journalist Thomas Goorden.

I find it a bit depressing how quickly colleague journalists “call an expert” whenever they encounter even the most basic physics/chemistry calculation. My suspicion is that making mistakes with units is one of the big fears. It doesn't have to be that way!

datafrosch.fun/blog/unit-co...

Boven is onder, oorlog is vrede en... “AI binnen overheden in Vlaanderen is democratisch, betrouwbaar, mensgericht en duurzaam, met correct gebruik en beheer van data en toegepast met kennis van zaken.”

It does get the “taken out of context” disinformation price of the day.

So, for anyone confused (or suffering from a rage boner): This seems to be from an innocuous/silly advice column from THREE years ago, not some kind of tragic downfall due to that billionaire.

AI is a hammer and the stock market thinks everything on earth is a nail and they’re on the cusp of realizing that this is far from true

Pondcast #3: Calculate CO2 emissions with Thomas Goorden

🧪 You don't need a PhD to fact-check emissions numbers.

@goorden.be (investigative journalist, physics background and a wonderful teacher) showed us at The Pond how three numbers from the periodic table let you sniff out bullshit in climate reporting.

Full session on Youtube: youtu.be/qE9GMVyRoEI

“Fly me to the moon, but don't be sycopanthic about it.”

AGAIN?!?

Dat is één manier om het te zeggen...

I'd also suggest “regression to the mean”, which might be the root cause for confabulation. But yeah, I also doubt it can be solved. On the other hand, they *can* get it good enough to fool some people. But due to point 2, that only works if you lie about its nature.

Maar dat is het rare hieraan: Eigenlijk weten we toch al lang waar dit vandaan komt?

"De resultaten bevestigen het vermoeden dat PFAS ook via neerslag in de bodem terecht kan komen"

Dat is pertinente zever. Er is helemaal geen vermoeden, we weten dat al lang.

De bron van “vervuiling in regenwater” is in feite altijd een luchtemissie hé. De vraag is of het puntemissies zijn (fabriek), of iets anders.

I believe there are two factors here:
1) The “sloppiness” of gen-AI, so basically a quality issue.
2) Our rejection of gen-AI as being culturally and thus economically valuable.

Point 1 is solvable in theory (although scepticism is warranted), point 2 is pretty much unsolvable.

Het rare is dat we al heel lang weten dat PFAS via lucht verspreid geraakt. Er zijn zelfs luchtmodelleringen gemaakt van de Chemours uitstoot in... 2017.

“De stoffen PFBA en TFA werden op bijna alle locaties gevonden.”

TFA is in ieder geval een rechtstreeks afbraakproduct van de uitstoot van Chemours, plus van bepaalde koelgassen. PFBA is wss ook industrieel van oorsprong.

PFOS is opvallender, dat is wss hercirculatie. In België ook gezien.

Are you saying I have (good) friends because I'm quite poor? 😅

(That's a bit funny, but I can also see the truth in this study.)

Eigenhandig een nieuwe generatie van protestanten uitlokken, ge moet het Rik Torfs nageven!