dimask

It should be enough to be cheaper if you include the environmental cost that fossil fuels cause to society and planet.

Video tl;dr:

Psychtoolbox is a popular open source tool for creating vision and neuroscience experiments. As typical of open source projects, the project mostly runs by a single, underpaid developer's time (Mario Kleiner). They tried to be funded through "support membership" donations, it did not work, and now they announce that a paid license will be required for running their compiled binaries "on some operations systems".

The relevant part where they describe the funding situation of the project starts around 6:00 as in https://youtu.be/05gpkP_EMoc?t=363 (sorry for messing up the link in the post)

In a chain of thought manner, as every proper AI, of course.

I think it is good to bear in mind that the distinction between perception, cognition and action/behaviour is in reality a reduction. There is more to these categories about the way we communicate and organise our society rather than mapping to a corresponding reality.

We cannot really handle incogruencies between these domains. The meanings we give to things, situations, people are not merely thoughts, they are stuff we enact on them. We can perceive something as unfair or destructive, but if nothing can be done to change it (or what can be done has too high a cost so it is practically inaccessible) we will tend to seek whatever sort of congruence we can. It is hard to think one way and act another, because that requires more energy which we are evolved to try to minimise. Hard realisations are that hard because they imply a sharp change in behaviour and way of living that we may not want, and because the implications span many domains that need to change. Author's family realising the reality of the mother would require changes in the perception of the family situation that was too uncomfortable. It is easier to hide things under the carpet to preserve the facade of a normal family life, than enact all the changes required. This is why such situations are so common when it is about families.

So, I do not quite agree with the premise and conclusion of the article, in the sense that I do not think that there is anything ingrained in our primate brains regarding how we judge incoming negative information, but rather something ingrained in our cultural institutions and social reality, which is what makes them both stable and dysfunctional. Seemingly irrational behaviours make perfect sense once one can understand how to contextualise them.

> support for running [...] software which doesn’t play nicely with major OS changes every few months

Half-kidding, this is a normal experience for people with macbooks coming from macos, ie updating the os and praying that all software will be working properly after. So it is probably not such a big deal in this case.

Imo if the northern lights are strong, the live experience is superior to photos because of how they move, while photos are still images. But many times they are not as strong and they do not look like much with naked eye, while long exposure will catch more of it.

The target audience is neither engineers nor general public; these announcements are meant for tech journalists/youtubers etc to refer to when writing or talking about it.

A VC calling out BS jobs in big tech, oh the irony.

> The things where computers shine are actually managimg the complexity of a large and complicated systems.

I would argue that where computers shine, firstmost, is automating repeated tasks. Even if a task is fairly simple in complexity and doing it by hand takes less time, if you have to repeat the same task over and over it may be beneficial to use computer tools that allow some automation, even if in the first couple of runs this is gonna take more time. In this sense, something being easier to do by hand (once) does not necessarily imply that it is better to do it by hand.

But I do agree that an increase in complexity comes as a curse of tools. People with less tech-understanding may be more easy to get what some benefits of such tools are, but the problems that increased complexity brings takes longer to catch them.

> Can this be combined with quantization?

It is in their TODO part in https://github.com/Infini-AI-Lab/Sequoia/tree/main

It is not about "infinite number of combinations", it is about browsers having some specific functionality as of now (not just in iOS), people preferring using web versions of apps on them because of that instead of installing apps, and app developers seeing that people use the web versions and saying "oh people are using the web versions, maybe we should give these the same functionality as the apps", disregarding that people _choose_ not to install their (crappy) apps exactly because of these "features" and "functionality", which usually mostly are about sucking your data and spamming you with ads.

All this apple-related HN web-browser drama is not about what _users_ want, it is only about what app developers and companies want.

> you can be doing something else in that 90 seconds, can't you?

If I get distracted from my task, not really, or at least not as focused as I was before the "hi". Except if it is something like the beginning of the day, where it does not matter much.

> You don't want that, but people are looking at gmail or google calendar on the web, and notifications are a huge QOL.

I look at the emails/calendar on the web _because_ I do not want notifications and other junk/distractions. I do not think it is big deal for people to have an email client app on their phone if they want their functionality.

If it is a company laptop, it can be impossible (unless you sort of hack it to circumvent the security settings they put).

Thank you for sharing this! Looking forward to try it, seems like an interesting idea. My strategy on bandcamp is conceptually similar, albeit manually.

To be fair, while they did introduce new restrictions in the past, for example disabling imap support for free email accounts, the restrictions did not apply to older accounts that had already these features activated. Ie if you were using imap before the change, you would not have to pay to continue using it.

Ime zoho has been very reliable. Not relying on venture capital [0] and seemingly working on longer than shorter term profit makes it imo also less likely to resort to this type of strategies.

[0] https://techcrunch.com/2022/09/10/how-zoho-became-1b-company...

> If Meta starts being all open and generous about their core assets

Didn't React come from them? PyTorch?

Making things open is a means, not an end, _in general_. The end is what we do with these things. What you call "competition" I would call "monopoly". A healthy, non-monopolic (or such) state of AI would be open by itself.

I would bet for this use case LLMs to be worse than random walk, as they would probably disregard certain possibilities in crucial steps, unless we are talking about easier stuff.

> I have 10 apples. I find 3 gold coins in the bottom of a river. The river runs near a big city that has something to do with what I can spend the coins on. I then lose 4 apples but gain a gold coin. Three birds run into my path and drop 6 apples each. I play an online game and win 6 gold coins but I have to share them equally with my 2 teammates. I buy apples for all the coins I have. The price of an apple is 0.5 coins. How many apples do I have? And where is the river?

> You now have 34 apples!

Actually, it forgot to add one coin when the person lost 4 apples, and the apples should be 34+2=36 in the end, so the model got it wrong. I tried on phind.com with their 70B model and it actually got it correct. Still, quite impressive for a 8B model, but it does show again the problems that come with trusting such problems to LLMs, that actually finding mistakes is so hard because it all sounds correct and confidently well written.

And quantised gguf files by Bartowski if somebody wants to download and run through llama.cpp directly

https://huggingface.co/bartowski/WizardLM-2-7B-GGUF

Considering these models are released by "Microsoft AI", I doubt they do anything against the ToS of "Open"AI.

Apart from what the article explains for TB3 vs USB4 enclosures, there are several additional issues with what connects how in an apple silicon mac. It is a mess to figure out an optimal set of enclosures and ssds for a macbook, as some combinations work better than other in apple silicon, while not having any difference in intel macs. For example, a lot of usb3.1 gen 2 enclosures seem to be recognised as gen1 (5Gb/s vs 10Gb/s). I don't think it is very clear why, but probably having to do with the specific usb controllers. In general, TB controllers seem to be more reliable.

https://forums.macrumors.com/threads/usb-on-m1-macs-isnt-act...

There are a lot of other factors to consider too, like how many pci-e lanes is dedicated in a usb port, how they may be shared between usb ports etc, eg it seems that some usb ports that are usb3.1 gen 2 are actually limited to ~8Gb/s because of that

https://forums.macrumors.com/threads/issues-with-usb-ports-o...

Advice: get an enclosure and ssd that you have seen somewhere in the internet as working with your device.

> for the Intel Macs it does not make much of a difference

Not exactly, as it actually mentions intel macs do not support USB4, so USB4 enclosures will fall back to USB3 which is slower than TB3. The article is no less confusing than the actual topic. So the actual advice is TB3 for intel macs, USB4 for apple silicon.

If the title was "meditation on math errors" it would have been a perfect art piece.

No, but if you wanted to count how many coconuts you had, you would start with 0 (pointing at nothing) and then you would continue by 1,2 etc pointing to each other coconut. This would allow you to use the same counting process to count something, even if there was nothing to count.

And? What are they gonna do about it (apart from making such a person/website momentarily famous).

Actually it seems like a feature to me (to be able to exchange encrypted emails so easily between different providers), and a quite nice one actually.

I doubt that 10xGPT3.5 > GPT4. There are a lot of tasks that GPT4 can do and GPT3.5 just cannot. Also, in such cases I find that GPT3.5's hallucinations are quite consistent, so such a method is probably not gonna help.

Same can go for electricity, water (if there is no local enough source of good water), as well as transportation infrastructure, mobile network coverage, and access to goods. I guess the main question is, does it make sense for a lot of such places to be abandoned, just because they will never be able to compete with cities in terms of cost due to distance and being smaller markets?