LLMs might theoretically give 10x senior devs 100x potential, were it not for the limit of the brain. The low hanging fruit is all the drudge work, all the stuff that you can envision exactly how it will look when done. But instead of all the fun problem solving and typing, you turn into solely an architect / pull request reviewer / tester role. You still have to actually read the code and wrap your head around it, and check every little aspect of it. At a certain point you get exhausted and it is way before 100x. Exactly how many "x" would be interesting to document, based on someone who is doing a responsible job. Not unrelated to the "30 hours/week max of creative work" quote in the 4 day workweek thread. Personally I am working less hours but every day of the week in order to spread out the sheer amount of information processing and eyestrain, which I find are the limiting factors. YMMV.

Something that bugged me. My (major) bank issued me both credit card and debit card. They counseled me to only use the credit card because it has fraud protection. And recently now I got a $10 fraudulent charge against my card. If the bank is issuing both cards and I am using it well within my balance for recurring billing and the like, why can't they cover both? When I called the credit card people they said the bank covers it. It's like it is the same people..

Agreed. Also, my iPhone 11 Max's Siri while not too intelligent is near-perfect at doing simple things like turning on and off a reminder or alarm, or the flashlight, when I am half asleep. I can tell it to set a 10 minute timer and know it is just going to work. (A few times it has said nothing but interacted without voice, so I would say just 99% perfect). Not so much on email search functionality, but the Search itself is not great. Also would like to be able to play a YouTube Music song without unlocking it, if it is night and the phone is not moving. But you know what? Bulletproof and secure is 100% a win for me. I don't want my info being sent to some other provider and stored forever. I don't know any other company I could trust to do that.

Agreed more interested in something with arms and maybe legs.. even chopsticks and stilts would be okay tbh. Reachy has no arms, its antennae look like they'll poke someone's eye out when they suddenly twitch, and listening/staring at your keyboard is probably not good from a security perspective.. that said, kudos to launching something. Maybe some limbs will be sold on a later reachy? I'm guessing it is for kids and maybe good for elderly like emotional support robot (it's a thing) that can call for help.

tldr; google Ai overview queried
how delicious is ostrich or emu compared to the tinamou?
(moa is supposed to be related to tinamou.)
tldr answer: should be like a delicate quail or pheasant!
but you could just run a tinamou farm, though not as cool!

Ostrich:
Taste: Ostrich meat is often described as having a taste similar to beef or veal, with some describing a subtle butter or toasted walnut flavor.
Texture: It's known for being tender and juicy, especially when prepared correctly.
Comparisons: Many find it a good substitute for beef in various dishes.
Emu:
Taste: Emu meat is often described as having a more pronounced, slightly gamier taste compared to ostrich, which some find appealing.
Texture: It can be slightly less tender than ostrich, but still suitable for various cooking methods.
Comparisons: Some find it similar to bison or venison.
Tinamou:
Taste: Tinamou has a more delicate, refined flavor, often compared to pheasant or grouse.
Texture: It's generally described as tender and fine-textured.
Comparisons: Its subtle flavor profile differs significantly from the richer, beef-like flavors of ostrich and emu.
In essence: If you're looking for a red meat with a flavor profile similar to beef, ostrich is often preferred. Emu offers a similar experience but with a potentially stronger, gamier flavor. Tinamou, on the other hand, provides a more delicate and subtle flavor profile, distinct from both ostrich and emu.

...They had to show they delivered something new using Apple Intelligence.
Can't tell if it is actually using the onboard neural engine or is it going through them.

I wondered if Xerox would get the keyboards but apparently Unicomp bought buckling spring from Lexmark in 1996 and now sells New Model M with buckling spring tech. Also a Mac version.. hmm wonder how it compares to keychron / MX keys?
https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.igorslab.de%2Fen%2Funi...
https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.pckeyboard.com%2Fpag...

Made it to the BBC too. tldr, sounds great but:
1. They apparently used puzzle cases that would be hard for humans. Is AI more on a par with humans for non-puzzle cases?
2. How sure are we that the quoted models have not already been trained on those puzzle cases, or other cases perhaps in medical school exams that were created based on knowledge of them? It sounds pretty suspicious for such a disparity.
3. If doctor sees two potential solutions with similar probability (maybe 60% vs 70%) they might pick either one, but a computer likely would pick the more probable one. What was the reason humans were bad at it. Was it just tons of data they could not crunch and so they got fooled by the most obvious points which (being a puzzle problem) tricked them as planned?
4. So how do doctors usually handle difficult cases IRL? IANAD but my guess is something like: Local doctors don't know and (possibly after lots of consultations and time lost with wrong treatments) send the patient to a top tier medical research facility where the world's top experts are. More tests are implemented to narrow down possible diagnoses. Then I am guessing treatments are executed and as they fail other treatments are given. Maybe a researcher picks them up for a new study, etc. An LLM doesn't actually care about healing the sick, it is just a math problem. Humans I am guessing are going to try to heal them by trying different things and I am guessing this is actually how it works. It isn't clear the AI will actually deliver the best results in the end, with the exception that it saves a lot of wasted time and costs (unless it ends up costing a lot to use an AI, because medical industry and liability).

Long time resident, have recently asked Claude to check my translations and flag any bad problems. It is really good at catching typos I missed even when reviewing when I'm tired and has recommended grammar or stylistic changes that make it sound better. But this is just writing technical docs, not prose! It is good at telling me what is more common usage (which is great since obviously it knows common usage really well) and can tell me why. Actually it is really good as a living dictionary and never gets tired ;) But.. never trust Claude to translate a spreadsheet without skipping rows. I made a pretty tool with it but only after realizing how cavalier it could be, saying it had checked things it didn't, etc.

You could ask it for something a lot simpler. Before reading this thread I happened to ask Claude to add a bit where a map of userid to record counts was outputted sorted by count in descending order. I was just kind of tired and ordinary Comparison didn't work in the dev console. Guess what Claude did? It made an array of counts, sorted THAT, then USED THAT AS A KEY to get the user ids. After I asked what on earth caused you to do such a silly thing it gave me a bubble sort. Okay fine, I don't actually depend on Claude for anything important but that was just beyond brain dead into WTF territory. Actual chat below:

This approach:

Creates a reverse mapping from count to list of user IDs
Sorts the counts in ascending order
Reverses the list to get descending order
Prints results grouped by count (highest to lowest)

Retry

Um.. I think you are doing something very dumb here. You have a list of counts and sorting it. And then you use the count as a key to get the user name. But there will be many users with the same number of records! Count is not unique so it cannot be used as a key! Where on earth did you think up such a silly thing.

You're absolutely right - that was a terrible approach! I was overthinking this completely. Here's the simple, correct solution:

tldr; Could be just me but a company with this much money can afford to issue security patches at least. I know I'll have a new MBP soon and only up to Monterey works on this one but still.

I was once entranced with a French experiment that made digital analogs of the paper ATC strips, called Digistrips. It is *still* way cool. Found it again. Ah yes! It was Zinc, and TkZinc. Maybe someone should ask other countries what *they* use? Or what the navy uses.. something tells me they are digitized already. Anyway the pretty links for your reference.
https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fgithub.com%2Fasb-capfan%2F...
http://doro.poltava.ua/tkzinc/...
https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.researchgate.net%2Fp...
https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.researchgate.net%2Ff...
https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.researchgate.net%2Fs...

Apparently it breaks down to a food additive and a fertilizer..

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.riken.jp%2Fen%2Fnews_p...
After screening various molecules, the team found that a combination of sodium hexametaphosphate (a common food additive) and guanidinium ion-based monomers (used for fertilizers and soil conditioners) formed ‘salt bridges’ that bind the compounds together with strong cross-linked bonds. These types of bonds serve as the ‘lock’, providing the material with strength and flexibility, explains Aida.
“Screening molecules can be like looking for a needle in a haystack,” he says. “But we found the combination early on, which made us think, ‘This could actually work’.”
In their study, Aida’s team produced a small sheet of this supramolecular material by mixing the compounds in water. The solution separated into two layers, the bottom viscous and the top watery, a spontaneous reaction that surprised the team. The viscous bottom layer contained the compounds bound with salt bridges. This layer was extracted and dried to create a plastic-like sheet.
The sheet was not only as strong as conventional plastics, but also non-flammable, colorless and transparent, giving it great versatility. Importantly, the sheets degraded back into raw materials when soaked in salt water, as the electrolytes in the salt water opened the salt bridge ‘locks’. The team’s experiments showed that their sheets disintegrated in salt water after 8 and a half hours.
The sheet can also be made waterproof with a hydrophobic coating. Even when waterproofed, the team found that the material can dissolve just as quickly as non-coated sheets if its surface is scratched to allow the salt to penetrate, says Aida.

https://ancillary-proxy.atarimworker.io?url=https%3A%2F%2Fwww.science.org%2Fdoi%2F10...

Editor’s summary

A strong, glassy supramolecular polymer has been shown to prevent the formation of marine microplastics by slowly dissolving in salt water into metabolizable compounds. Cheng et al. show that salt bridging between sodium hexametaphosphate or sulfated polysaccharides and guanidinium sulfates expels sodium sulfate to create a cross-linked network that is stable until the electrolytes are added back. The dried material is a moldable and recyclable thermoplastic that can be water stabilized with hydrophobic coatings. —Phil Szuromi
Abstract

Plastics that can metabolize in oceans are highly sought for a sustainable future. In this work, we report the noncovalent synthesis of unprecedented plastics that are mechanically strong yet metabolizable under biologically relevant conditions owing to their dissociative nature with electrolytes. Salt-bridging sodium hexametaphosphate with di- or tritopic guanidinium sulfate in water forms a cross-linked supramolecular network, which is stable unless electrolytes are resupplied. This unusual stability is caused by a liquid-liquid phase separation that expels sodium sulfate, generated upon salt bridging, into a water-rich phase. Drying the remaining condensed liquid phase yields glassy plastics that are thermally reshapable, such as thermoplastics, and usable even in aqueous media with hydrophobic parylene C coating. This approach can be extended to polysaccharide-based supramolecular plastics that are applicable for three-dimensional printing.

I worked at a big bank and knowledge management was an after-thought. Confluence was seen as a "A place where knowledge goes to die". I heard about StackOverflow for teams and thought it was a good fit -- we have a lot of siloed (even worse with Covid/WFH) developers and a lot of large complex systems. Just knowing who to ask was often a mystery. Having a place to ask and share would have been a great tool.

Unfortunately management churn ensured it never made it to the funding stage, and then co-pilot (which totally lacked our institutional knowledge) was posited as an alternative. That was the end of that. Sad.

Actually if you read the paper it is more interesting than that. Different models are tested at running a company and they variously do well, fail, recover, etc. When they fail and lose track of say orders (which I am assuming means the whole system is way too open-ended and badly designed) they try to escalate. Then the person running this thing just says "continue on your mission". This is the clincher. A human in the same situation, if they were high maybe, could have the same reaction as the freaked out agent that asks for some kind of advice instead of being locked in a prison without guidance (funny / not-funny). When it veers, it veers strongly and yes gets totally unhinged calling the FBI or using crazy talk about laws of the universe, although it is saying "It is not physically possible to continue my mission with the company closed down, idiot!" It just doesn't know any swear words. So yeah they talk about long running models all losing coherence over time, but as they have totally unsanitized I/o, a sadistic over-manager pressuring the agent (and we know what happens then, shades of threatening your LLM), and have no negative feedback loop (at least none I saw in my cursory reading) that could bring them back to sanity, this looks like a bit of an artificially tilted study. The interesting part was that different success at gaming was seen, scored on net worth. The failures so far seem to be coming from 1) weak overall structure 2) lack of sanity checks 3) lack of positively reinforcing guidance i.e. from a boss or board of directors that is sane and 4) random mistakes and chaos which we already know LLMs do. The posited "loss of coherence over time" might be true but it seems more like classic thermodynamics and not just because LLM.