flowerthoughts

Nice! Thanks for posting. How critical is the 1.28s chunking? Is that to ensure getting a full word? I'd think getting final-word latency down would be a goal. Does the 0.9s processing delay scale down with the sample size?

What was wrong with ML? How many terms do we need to churn? What happened to AGI? Is that SGI now?

My reason is their revenue chart is flattening.

https://techstartups.com/2026/10/15/anthropics-q3-revenue-hi...

> It's not deblurring, there's no post processing. It just selects where to autofocus, and commands the lens to focus to that point, that's all.

Ah, thanks for the clarification.

I could imagine the EU would consider banning that for privacy reasons if it's raised by the right organization.

As an avid reader: thanks for helping with that!

If the point of a competition is to show a photographer's skill, it sounds like providing deblurring or similar by educated guess would help quite a bit to win?

I guess the key point is that use of features needs to be adopted slowly in competition so that it doesn't turn into a materials sport where it's just the most expensive equipment winning.

AI photo editing is surely a competition-worthy art of its own.

Edit: moot point; it's object recognition to find the auto-focus target, not applying digital refocusing.

I've considered embedding an HSM in my concrete floor, with a USB-cable sticking out to plug into the server. :)

Assuming someone either steals your entire server or nothing, it's either that, password at boot, or some remote unlock (KSM). I doubt a burglar is going to just steal your harddrive. But I doubt they'll think of opening the floor.

Can you quantify the "harder to hack han a server"? This seems like one of those risks that is superficially true, but in practice is low enough to not matter.

Where are your E2EE keys?

Same reason they sponsor Firefox; it helps keep anti-competition lawsuits at bay?

If you cannot accept that a hair was on a bed sheet and it was fixed promptly when reported, I cannot even imagine what your hotel bills look like. Every hotel room I've been to at $100-$2000 per night have had some issue I would consider beyond a hair on a bed sheet.

I remember 10 years ago when there greentech was a thing and "carbon neutral" (by carbon credit shams or not) was a goal. I was a bit proud then. That has passed.

Now it's just about heating data centers as much as possible. Apparently climate is a solved problem. The tech CEO's children and grandchildren are safe again, and we can go back to focus on profits. /s

Tired of the complete inability of the human race to take in the Pale Blue Dot [1], and what it means. No one is coming to rescue your grandchildren. And it would be much easier for us to do something today, than them in 50 years. Actions compound over time: perhaps those actions should then be long-term positive.

1. https://en.wikipedia.org/wiki/Pale_Blue_Dot

> The next 1929 is going to happen faster than the last one

For sure, and for reasons we can't imagine today.

I sometimes think about nuclear accidents. It seems very unlikely that we've already seen the worst accident that will ever be. Same goes for so many technologies less than 100 years old. Safety standards need to go up in parity with the scaling up to keep MTBF in check. Given what I've seen of human greed, I hold only partial hope that'll happen as energy demands increase. Hopefully batteries, wind, solar and regulatory oversight are enough.

That's more about being easily ignited. It still requires a spark or heat source (like an oven) to combust.

> “it doesn’t feel safer.”

> This conversation is usually about how no one lets their kids play outside unsupervised anymore,

Those two show overall fear is higher, not that reasonable fear is higher.

There's so much in current society that sells fear, that I'm not surprised. FOMO for the next Musk stock promotion. Stalkers, scammers, pedophiles. Burglar alarms being sold with a subscription at 70% profit margins (Verisure here in Sweden.) I recently calculated that I'd pay about $300k-600k in fees for every actual burglary where I live. The insurance deductible is way way lower than that, and I doubt they'd actually change the outcome. Incidentally, insurance companies care about water and fire damage, not burglary. But the scary marketing video with masked men in your garage sells well...

Of course, insurance is a a whole other chapter on selling fear. Sometimes warranted, mostly not. The Freetrailer app, used to book and pay for trailers, defaults to me paying for a deductibility reduction. With a big fat warning if I remove that, that if something happens, I might be liable for up to $700!

Not to mention the economy is always just one bad quarter away from a 1929 crash.

I blame media (new and old) for the fearmongering that sells my attention on low probability outcomes.

I must be missing something.

1. If it costs $6-22 million to produce a paper that the brightest minds in maths can't understand and use, this trend will stop soon enough on its own.

2. Why are these minds complaining that there are no authors to discuss the paper with? There are OpenAI researchers to talk to. If they refuse, because they themselves don't understand, then don't bother with it? You're responsible for your time. Write a response on what's unclear, and move on. What's the point of whining about others writing things you don't understand? See point 1.

3. If LLMs can be used to solve problems, that leaves more time for mathematicians to come up with follow-ups. That seems a great way to speed up the field, no? Just be sure you understand the solution before expecting anyone else to spend time on reading it.

I know next to nothing about Bulgaria, other than where it is.

Is that sympathy expressed for the people or for the nation? I definitely have sympathy for Russian mothers, having their sons sent to war. And the emigrants that can barely return to meet their family. But I'd also be happy to fight those sons if they think Russia has claims to anything outside previously agreed borders.

Would Claude have built the thing without me, for me to use? If not, I'll take full credit.

Would I have built the thing without Claude? If not, I'll retain some credit and leave some credit to Anthropic's team for the help.

The inflation run in the seventies, ending in 1981, probably helped quite a bit in eroding trust. I'm going to call non-causation on Watergate.

> This year’s excavation unearthed 131 cuneiform tablets and fragments which record an epidemic of fever, military logistics, religious practices and events of royal daily life 3,400 years ago.

Nice time machine you've got there. Not just vases and bones, but thoughts and societal complexities.

That's relieving to hear. Thanks.

That's darn nice. I didn't even have to push the turbo button to be able to run it on my phone.

I wholeheartedly bless this slop.

Found the repo in the Reddit post:

https://github.com/terrapapagalli1516/quake-srp

https://www.reddit.com/r/quake/comments/1x1ch14/quake_srp_sl...

Who am I to dump on Terence Tao, but

> Mathematicians have a particular vision of progress that is informed by history and field-specific considerations.

really sounds like something a side-quest association would produce in a panic response to someone trying something different. It's just an _ad hominem_ and gatekeeping argument.

That was a pointless source. Which 23 categories? How do I know if I care about them?

Fig. 2 [1] lists them. Fig. 3 I think indicates that 20 of them have positive effect size on biodiversity, which I think was the point. "Learning and inspiration" is an interesting category. Carbon sequestration ranks high, with Terrestrial (as opposed to the larger Oceanic) having quite a few studies behind it.

1. https://www.nature.com/articles/s41559-026-03200-4

Possession of CSAM was not a crime until it gave the state a reason to go through every datum you own. When the incentives are right, laws can become anything.

In Europe, we suddenly had laws against "hate speech," which I personally think goes against the Swedish constitution's protection of free speech, but here we are.

I'm gonna go invest in some psycotherapy companies, because the world is gonna need to repair a lot of trauma after the artificial intelligences that currently rule the headlines are done.

Oh, agreed this sounds like a terrible migration path and shouldn't really be needed in the first place.

What I'm missing in the article is whether any Git server accepts replacing a SHA-1 identified object it already has. If it doesn't, then the distribution trust discussed holds, and keeping SHA-1 seems fine. Adding additional signatures seems fine for those who need transitive trust.

The news is that lethal injections are legal, but required a retry. GP asked: So why not a firing squad? Them retrying might be less cruel than reinjection.

It would be funny if the solution is for Greenland/Denmark to annex Canada as a workaround on paper.

Proposed new rule:

If there's no technical reason to fuck with the product I have already purchased access to, don't fuck with the product I have already received.

The nice thing about others taking photos, is that I can focus on the odd ones. I have photos of really well made knots on a guiding rope, and street gutter stones.

You wanna see what Venice looks like in general? Just go on Tiktok; they have better equipment anyway. You wanna see, up close, the nicely carved gutter stones that handles water without breaking the stone pattern? I'm your photographer.