The father-mother-child-donkey conundrum. Its only solution is: to hell with anyone's thoughts, we'll act as we prefer.

I had to look this up [1]:

> C. T. Studd, (1860-1931) the famed English missionary who served God in China, India, and Africa, once remarked: "Remember the miller's donkey. The miller, son and donkey went to the market. The miller rode the donkey all the way and people exclaimed, 'Cruel man, riding himself and making his son walk.' So he got down and his son rode; then people slanged, 'What a lazy son for riding while poor old father walks.' Then both father and son rode, and people then said, 'Cruelty to animals, poor donkey.' So they got down and carried the donkey on a pole, but folks said, 'Here are two asses carrying another ass.' Then all three walked and people said, 'What fools to have a donkey and not ride it.' So let's go ahead with our work for God and not care what folks say."

Secular wisdom for the ages! 150 years ago we would have been taking a donkey to market wrong. Now with the advent of YouTube and Reddit and Twitter we've charted vast new swaths of incorrect behavior that we can leverage to make ourselves feel superior to others.

[1] https://www.actsweb.org/articles/article.php?i=295&d=2&c=6

Because egress firewalls have existed since way before "ai" and are very easy to set up.

But that’s an objection that OpenAI should take some easy action, the framing that OpenAI has out of control agents is still true.

Seems you think there is a silent “…and there is nothing they can do about it” after “OpenAI has rogue agents”?

The framing is important. Agency and responsibility lies with the humans at OpenAI, not with the bots.

If your kid steals your car, punish them and try to prevent it from happening again. If your kid steals your car a half-dozen times, crashing through a storefront each time, and you still leave the keys out, the story changes. At that point, negligence becomes complicity.

Yes, of course, how is any of this incompatible with OpenAI having out-of-control agents?

Agents are out of control by default, especially during a training run, which is why when they are productionised into cloud hosts or even local harnesses they have external guardrails in place

In other words, I have a gun that shoots bullets. It's up to me to use it responsibly and legally.

And memorizing. Rote learning makes the brain.

Not sure why this was downvoted. Memory is one of the most important skills imo

Having a good memory enables you to think deeper - if you remember certain concepts, you can reason better because you remember more concepts so your thinking is more in depth.

More things in memory also leads to making more connections / recognizing more types of patterns more easily.

There are things that you just see/know/recognize that are not easily searchable/promptable because the fact that you should be searching/prompting would not occur if you had not seen the pattern before.

One could argue "experience" is just the accumulation of memory such that you can recognize more patterns more rapidly, short circuiting bad decisions before they are made. With LLMs its even more important to have a well tuned "what should be getting built" mental framework if that becomes our primary job.

There's memorising and there's memorising.

My history lessons at school were entirely "here's 30 dates of these important events happening, there will be a test at the end of the month where you have to name the event at each of the given dates". I just remember sitting there trying to somehow remember what happened on each of those random dates, what event was it that I knew nothing about with zero context other than the fact that it happened. I can confidently say that after I ended my formal education I remember 0 of any of those dates. That approach to teaching history is just completely wrong, I would have loved to be told WHY certain things happened not just when they happened, but the system wasn't designed for it.

But maths formulas or even chemistry and physics - yes, that makes a lot of sense because you need to remember them and then you can actually use them to solve problems later.

Yep. Even for maths, knowing why is king. Once you understand maths there is a whole bunch that makes sense and you can just derive.

You were downvoted but also right. Memorization is an extremely useful skill to develop.

Rote memorization is very ineffective way to memorize things. That is why it was downvoted.

Sorry what? Honest question.

Microsoft allows you to subscribe to Microsoft 365 personal or family without AI for a small discount, but they hid the option on a page you can only get to if you click cancel on your subscription.

Microsoft are currently being sued in Australia by consumer protection (ACCC) over this practice.

Iirc Microsoft increased the pricing on existing subscriptions on next renewal citing AI as the reason for the increase, they then added a no AI version at the original price, effectively opting all users into the 'upgrade' rather than requiring them to explicitly upgrade their licence.

[un]believable. Thank you.

H..how is this legal.. in 2026

Likely illegal in the EU. But in the US? 2026 is the year of open corruption, as long as the administration likes you you can do whatever you want.

It’s really surprising to me that after the US public hasn’t internalized how corrupt the country has become. It’s all coming from the top

Some of us have internalized it! It's like walking around in the twilight zone or the bodysnatchers movie sometimes. In a dream society we could agree on some values and argue about how to direct our shared efforts. In this one we can't agree on values or even facts. It's equal parts maddening and terrifying.

We need help in avoiding the worst futures that could come from this.

The US has never been strong on consumer protection. It’s that individualism and buyer beware culture.

Nah, not illegal in the EU. Giving a retention offer to leavers is something magazines sometimes do. It's a bit sleazy, but not illegal.

If you have enough money nobody can tell you what you do.

It’s a documented fact that a lot of subscription services will offer you a discount if you ask to cancel. This has been a thing in the US with cable tv for about a decade, as far as I’m aware. I think some cell providers might do the same. Not everything, but some.

But in this case, the discount comes with a downgrade. Still worth it for many people, and Microsoft shouldn't hide it.

That is exactly what "training AI" needs: labeled data.

It might be that the velocity at the spiraling axis is not infinity (i.e. there is no blow-up in finite time). Viscosity is a bitch.

I believe a core element of the smoothness problem solution is that it does blow up in finite time, even with viscosity.

Perhaps an expert could provide a more accurate response, but here is how ChatGPT tried to explain it to me: "In this construction, the viscous term becomes large along with the acceleration, pressure-gradient, and nonlinear momentum-transfer terms, but they balance/cancel in a very precise way. That allows the velocity in the concentrating vortex core to grow without bound even though ν is nonzero."

That has another name but I forget. Essentially contour plotting in high res.

Do you mean Marching Squares? Or Signed Distance Fields? Both are related, but not exactly the same.

Level sets?

Yep: the problem with IP is exactly the "Property". The texts are economically (even if only for leisure) valuable, so the producer deserves a fair comoensation unless he has explicitly dismissed it.

So paying a price for an intelectual product is reasonable. Nothing to do (per se) with property.

And if attribution is their coin then they should be properly attributed even if they aren't paid.

Wow, this is really interesting. Thanks.

Apart from the well-known dubious position of OpenAI wrt truth, the prompts/inputs do mot include the outputs.

You can train on a sequence of outputs. In the end, OpenAI outputs are OpenAI's property.

You can learn a lot from a single side of a conversation.

But isn’t Tristan’s breakthrough happens in August? OpenAI can’t really train with text that doesn’t exist

But using the outputs to train would make their statement false, since they are influenced by the inputs

There is potentially a world of difference between how you interpret what is fair and what the terms of service contractually guarantee.