csdvrx

Will the offline mode work on laptops?

I use both: they are extremely fast - faster than Wordpad (on a native windows installation), much faster than libreoffice (on linux with wine).

Read https://blogs.windows.com/windowsexperience/2024/12/06/phi-s...

I submitted it, as it gives a better picture of what Microsoft is trying to do: both the hardware, and the software.

Phi is small, not just for shows, but also to be able to run locally on the hardware they are planning for it to run on the copilot branded devices.

Yes: they are building both the software and the hardware for that: https://blogs.windows.com/windowsexperience/2024/12/06/phi-s...

Is anyone here using phi-4 multimodal for image-to-text tasks?

The phi models often punch above their weight, and I got curious about the vision models after reading https://unsloth.ai/blog/phi4 stories of finetuning

Since lmarena.ai only has the phi-4 text model, I've tried "phi-4 multimodal instruct" from openrouter.ai.

However, the results I get are far below what I would have expected.

Is there any "Microsoft validated" source (like https://chat.qwen.ai/c/guest for qwen) to easily try phi4 vision?

Sometimes, you can't find the average because it's undefined: it can happen with a Cauchy and a few other statistical distributions: the wikipedia page has a nice plot of how the first 2 moments don't converge https://en.wikipedia.org/wiki/Cauchy_distribution#History

When in doubt, don't use the mean: prefer more robust estimates, as even with degenerate statistical distributions, there are still some "good numbers to report" like the mode or the median.

And if you don't know statistics, just use a plot!

it's like systemd trading off non-determinism for boot speed, when it takes 5 minutes to get through the POST

That's a bad analogy: if a given deterministic service ordering is needed for a service to correctly start (say because it doesn't start with the systemd unit), it means the non-deterministic systemd service units are not properly encoding the dependencies tree in the Before= and After=

When done properly, both solutions should work the same. However, the solution properly encoding the dependency graph (instead of just projecting it on a 1-dimensional sequence of numbers) will be more flexible: it's the better solution, because it will give you more speed but also more flexibility: you can see the branches any leaf depends on, remove leaves as needed, then cull the useless branches. You could add determinism if you want, but why bother?

It's like using the dependencies of linux packages, and leaving the job of resolving them to package managers (apt, pacman...): you can then remove the useless packages which are no longer required.

Compare that to doing a `make install` of everything to /usr/local in a specific order, as specified by a script: when done properly, both solutions will work, but one solution is clearly better than the other as it encodes more finely the existing dependencies instead of projecting them to a sequence.

You can add determinism if you want to follow a sequence (ex: `apt-get install make` before adding gcc, then add cuda...), or you can use meta package like build-essentials, but being restricted to a sequence gains you nothing.

In the last discussion about such issues, someone linked https://www.abc.net.au/listen/programs/rearvision/the-dark-s... which is worth a read for past abuses:

Military authorities in California requested census data to identify the Japanese-American population. Then in 1942, president Franklin Roosevelt issued an executive order to authorise their removal.

Very interesting link!

Submitted!

run ZFS scrub (or equivalent) every 6 months

zfs in mirror mode offers redundancy at the block level but scrub requires plugging the device

All error correction has a limit. If too many errors build up, it becomes unrecoverable errors

There are software solutions. You can specify the redundancy you want.

For long term storage, if using a single media that you can't plug and scrub, I recommend par2 (https://en.wikipedia.org/wiki/Parchive?useskin=vector) over NTFS: there are many NTFS file recovery tools, and it shouldn't be too hard to roll your own solution to use the redundancy when a given sector can't be read

For long term storage, prefer hard drives (careful about CMR vs SMR)

If you have specific random IO high performance needs, you can either

- get a SLC drive like https://news.solidigm.com/en-WW/230095-introducing-the-solid...

- make one yourself by hacking the firmware: https://news.ycombinator.com/item?id=40405578

Be careful when you use something "exotic", and do not trust drives that are too recent to be fully tested: I learned my lesson for M2 2230 drives https://www.reddit.com/r/zfs/comments/17pztue/warning_you_ma... which seems validated by the large numbers of similar experiences like https://github.com/openzfs/zfs/discussions/14793

Cosmopolitan is neat hack, but it's not a viable multiplatform executable format, runtime sysytem, or distribution mechanism.

It's the best we have to ensure programs will be kept running. The multiple payloads are like a rosetta stone.

If I had 1 wish, I would make wine support sixel output + VNC to a localhost port + webgl or equivalent to extend this redundancy to GUIs

Something like WASM + WASI seems much more likely to fulfill this function in the future.

Time will tell, but given 2 binaries (Sun Java vs Windows i386) from the same time period, the one that was "much more likely to fulfill this function in the future" is much harder to use

p-code is a neat technology, but failing at what it was supposed to achieve

The dream has been realized with the release of cosmopolitan.

There's no reason we couldn't have a cross-platform minimal set of common utilities.

You would have to add support for the peripherals in the kernel, and have some kind of init system. You would also need a filesystem supported to boot the computer.

I was doing something similar for the entire OS a few years ago: cosmopolinux, a distribution of cosmopolitan binaries: https://github.com/csdvrx/cosmopolinux

My idea was to replace the WSL binaries to have a Linux distribution living on C:\, but that could also be booted baremetal if you didn't want to use Windows

I had to put together a multi stage init system for that: if you get the ISO, you can put in on a thumbdrive and boot it: https://gitlab.com/csdvrx/cosmopolinux

The only difference between them is the kernel and the filesystem: the github NTFS has a firecracker linux kernel, the gitlab ISO has a regular kernel with many modules.

I wanted to do a full NTFS solution but I couldn't find a bootloader I liked that would support booting from a NTFS partition.

Booting from an ISO was simpler and faster.

If you find that interesting, read about the African Humid Period: https://en.wikipedia.org/wiki/African_humid_period

The present day situation is fascinating: https://en.wikipedia.org/wiki/African_humid_period#Present-d... : there is an ongoing "greening" which seems caused by global warming and CO2 increases!

However, a 2003 study estimated only 45% of the Sahara could be covered by vegetation, and a 2022 study found that it may not be sufficient to start another AHP: it just "lowers the threshold for orbital changes to induce Sahara greening"

You're going to get a frankenstein NEMA-Schuko-Chinese power outlet

They are very practical, as they accept any plug

That attack requires continuously monitoring a given device or area though, right?

The "randomization" seems to be a pseudo-randomization: with the seed and the timestamp, you should be able to deduce the future "randomized" addresses.

Also — secondhand books are generally outdated, undesirable, and/or damaged — do collectors still find diamonds in the rough?

Secondhand stores offer in general "better" products because of the double curation:

- someone found the good interesting enough to buy it in the first place (1st curation)

- the store found it good enough to buy it again from the first owner (2nd curation)

Aren't brick-and-mortar bookshops, generally speaking, as viable as Apatosaurus today?

No, it's even better because of the limited space they have to display the goods they want to sell: while online stores can show their full inventory, brick-and-mortar need to select what's most likely to sell.

This adds yet another level of curation: the store found the good valuable enough to be exposed to buyers, instead of keeping it in the back (3rd curation)

I find great music by randomly buying second-hand CDs from brick-and-mortar secondhand stores, thanks to this triple-curation,

The randomization doesn't matter: you can very easily link the addresses if you have a few datapoints, even if it's just the time you observed the addresses: the basic method is discussed in https://inria.hal.science/hal-03045555/document

See https://inria.hal.science/hal-02394629v1 for the theoretical bases then hop to https://samteplov.com/uploads/shmoocon20/slides.pdf for an example applying to Apple devices

Those who said the randomization and other techniques were sufficient were wrong: https://petsymposium.org/popets/2020/popets-2020-0003.pdf will show you how they changed their mind :)

It's not just apple: google nearby has also been reversed: https://publications.cispa.saarland/2748/ and https://www.ndss-symposium.org/wp-content/uploads/2019/02/nd... talks about attacks, but there's no need for that: just find identifiers that let you link the addresses

Even if you don't have any identifiers, the Bluetooth address randomization happens only about every 15 minutes: the manufacturer specific data in the public advertisement (or even the frequency and the length of these advertisements) during these 15 minutes periods can be used for linking the randomized addresses

And this is why I think gibson1 is wrong: we can argue about which projections or systems of logic should be used, concepts are still "calculations".

The gemini example was a wonderful summary of the comments, but audio is not very practical for something that long.

What about putting the text version that's used to make the audio somewhere on the page? (or better, on a subpage where there's no audio playback)

This approach works even better with a firecracker-style linux kernel (faster to start), and with cosmopolitan binaries (so you can do qemu tests not just with linux): a bash cosmopolitan binary is useful if you want to test your bash shell scripts in qemu

I have found the Gemini 2.5 Pro summary genuinely interesting: it adequately describes what I've read.

Have you thought about automatizing hn-summaries for say what the 5 top posts are at 8 AM EST?

That would be a simple product to test the market. If successful, it could be easily extended to a weekly newsletter summary.

My point is no type of math will work to model reason

Then I disagree with you.

IMHO interesting article that interacts with many debates on HN.

It's paywalled

A more generous take on the previous post is that the dominant paradigm of Math (consistent logic, which depends on many things like transitive preference) is wrong, and that another type of Math could work.

If you look at the slide, the subtree of correct answers exists, what's missing is just a way to make them more prevalent instead of less.

Personally, I think LeCun is just leaping to the wrong conclusion because he's sticking to the wrong tools for the job.

Intransitive preferences is well known to experimental economists, but a hard pill to swallow for many, as it destroys a lot of algorithms (which depends on that) and require more robust tools like https://en.wikipedia.org/wiki/Paraconsistent_logic

just one of the many tools of reason.

Read https://en.wikipedia.org/wiki/Preference_(economics)#Transit... then read https://pmc.ncbi.nlm.nih.gov/articles/PMC7058914/ and you will see there's a lot of data suggesting that indeed, it's just one of the many tools!

I think it's similar to how many dislike the non-deterministic output of LLM: when you use statistical tools, a non-deterministic output is a VERY nice feature to explore conceptual spaces with abductive reasoning: https://en.wikipedia.org/wiki/Abductive_reasoning

It's a tool I was using at a previous company, mixing LLMs, statistics and formal tools. I'm surprised there aren't more startups mixing LLM with z3 or even just prolog.

Returning to the topic of the limitations of LLMs, LeCun explains, "An LLM produces one token after another. It goes through a fixed amount of computation to produce a token, and that's clearly System 1—it's reactive, right? There's no reasoning," a reference to Daniel Kahneman's influential framework that distinguishes between the human brain's fast, intuitive method of thinking (System 1) and the method of slower, more deliberative reasoning (System 2).

Many people believe that "wants" come first, and are then followed by rationalizations. It's also a theory that's supported by medical imaging.

Maybe the LLM are a good emulation of system-2 (their perfomance sugggest it is), and what's missing is system-1, the "reptilian" brain, based on emotions like love, fear, aggression, (etc.).

For all we know, the system-1 could use the same embeddings, and just work in parallel and produce tokens that are used to guide the system-2.

Personally, I trust my "emotions" and "gut feelings": I believe they are things "not yet rationalized" by my system-2, coming straight from my system-1.

I know it's very unpopular among nerds, but it has worked well enough for me!

For albums, I recommend supraphonline.cz: https://www.supraphonline.cz/vyhledavani/alba/maingenre.1-ve...

That's fair!

Before using an existing solution, I like to understand how it works to make sure I will not get bad surprises: being able to reverse and debug using assembly code can be a helpful skill, and likewise for understanding DNS.

Yesterday I vibe code a DNS server from scratch in half a day, because I wanted to test something very specific bridging DNS and mDNS. Doing the same thing for health checks and geo routing may take what, another half day?

The experience and understanding gained can help decide if it's worth using a service like route53 or not, or even better: just doing without the feature, because if you have 1 VPS, "YAGNI" is the likely answer!

If the poster is seriously thinking about scaling to 2 VPS or more, the experience gained will expose the various ways it can fail, to maybe reconsider the decision (maybe instead get beefier hardware?)

In my case, I saw the DNS-mDNS isn't much a problem, so I don't have to reconsider adding the feature I want.