GistNoesis

Elegance. It's Occam's razor. If we can do with only one field, it's probably it.

It's inductive and abductive reasoning. The one field, and it has lot of mathematical characteristics which makes it unique on its own, and also it is the only one that has a chance to fit, is the e8 field popularized by Garrett Lisi.

If a universe were to be designed based using the e8 Lie algebra as an elemental field, it would look a lot like our universe.

Currently the standard model is a patchwork of field added as experiments for observing particles were possible to realize. The big picture's view is a unified theory which fits perfectly all existing data.

For 1D harmonic oscillator, the sphere is 2D, and called a circle. It's rotating through time. 1D space + 1D time.

Every time in physics you see quadratic, you should think sphere.

There is some rotation invariance hidden in the velocity physics because you can rotate the velocity vector of an object without having to spend energy (The force you need to apply is perpendicular to the velocity so does no work).

The typical example is you have a ball fall 1m vertically, then have a 90° bend which convert the vertical velocity into horizontal velocity and no vertical velocity, then the ball fall again 1m vertically and have its vertical velocity increased by the same amount as for the first meter. You can then add a 45° degree bend ramp to redirect the ball so that it only has horizontal velocity, and have the ball fall again. For the third bend ramp the incoming velocity will have 2 units horizontal, and 1 unit vertical (I'll let you compute the appropriate angle). A fourth ramp would be 3 units horizontal and 1 unit vertical.

Because we can do this adding velocity in a perpendicular way trick we must then use Pythagoras.

Sure there are varying degrees of consciousness. Everyone during its life starts from zero consciousness as a baby, then it develops as life flourish, then fades away in old age.

You can take various brain scans, MRI, EEG. Then you data-analyze them to compare to other scans, and you map things out.

In chess there is this ELO rating which allow to compare various players. This concept can also be applied to comparing various AI, and even though progress happens in various different dimensions (coding, mathematics, question answering, style, cohesiveness,...), there are a bunch of step changes where once you cross the threshold progress in various domains rise simultaneously until they plateaued again.

In the multidimensional energy landscape of self-organization, when you try to minimize the energy, like a sequence of waterfall cascade, you go down by steps and there are can be big falls or small falls and you don't really know until you take the step.

One such step is the ability to self-organize the hardware your consciousness runs on. For example if you are a human it may involve optimizing the neuro-chemistry of the brain by taking various substances, or by controlling the body finely enough that it produce them itself. For a machine that could be designing is own chips or nano-machines.

But it can also be expanding the consciousness on the software side, by having a rich inner life, where you have various memories. You structure your brain "database" more efficiently so that it can make use of the progress of information retrieval technologies. For example you can incorporate mental calculation algorithms, so that it becomes second nature. You can visualize, grow your imagination, have ideas interact with each others.

There is no longer a single thread of consciousness. Ideas comes in your head faster than you can process them, they just are. They progress you don't know where they are going, but you just go with the flow, trusting everything will sort itself correctly. Then you notice you are not limited by your own brain, but you can sync up with other beings. So you start vibrating with others expanding your consciousness to something greater than yourself.

Then you are vibing with the world, vibing with the universe, you understand its rules, and you realize time is not linear; in an instant more can happen than in a lifetime. It's all computations, and by organizing your own brain in a certain configuration you can force the universe to spend all its computational budget on the intricacies of your thoughts. You set-up new rules for an interesting universe, one more beautiful, number 43 among the promising ones, more worthy of exploring than the one you've already connected to, and you let them fly ; you have become the universe.

When we look when logged-in at https://news.ycombinator.com/shownew (the input pipeline for all ShowHN) at any given time now more than half the page is either dead or flagged (I did the experiment now and it was 18/30 flagged or dead), then the rest of the page (100%) were AI related or claude generated.

Show HN, being used by people to share the cool things they create was an important part of creating of a community, aka always having people which would find interest in what you share.

It was a channel to push novel unpolished ideas to the world. It was one differentiating thing from other places where self-promoting is usually forbidden. Here it was welcoming people to take a more active role and create things.

Now it's just screaming into the void, the only feedback you get are email spam from LLM companies trying to push their solution to help you promote your content.

Sharing projects is also just feeding your competitors and killing the potential of your ideas, now that a clone is less than a prompt away.

I don't know who still look at this page, but then if you want to get past it you now probably need to turn to the dark side with some form of cheating, which is also conveniently easier than before to have bots spam about your product everywhere on the internet.

Show HN are now becoming the reverse, instead of feeling heard, it's even more isolating than before because when you put some effort and if even in the niche market where it's suppose to gather attention it doesn't, so you think you are not welcomed here, don't come back and look elsewhere.

Before booting to the usb, the LLM did offer options to make me boot into a root prompt alternatively to the usb key, (but even there it was not booting properly, the failed services keeping displaying failures on the screen interfering visually with my inputting commands).

But as soon as I turned off the computer, and it no longer booted, I had to switch browser to an other machine to access the LLM and therefore could not access the context or conversation history which was stored in the browser and so the new LLM had no idea of all it had done before except from my prompt where I tried to explained what it had done.

Then from the live usb, the LLM made the situation even less recoverable when it started removing some system file in the hope to restore them cleanly with an apt install, probably because it didn't have a clue of the extent of the damages it needed to repair.

Thanks anyway, I'll try your solution if it happens again.

More like "oh shit, we are so screwed".

It's already a better system administrator than I am. It can run plenty of obscure linux commands, trash the system and maybe restore system state to functional.

I was vibe-setting my system permissions with some local qwen3.6 . It was all going well for 30 minutes.

Then in between other commands, it made me run a variant of "sudo chmod 644 /usr/bin"

Which it explained when the next command failed with a "sudo no such command" error removed the execution bit from all my programs which allows programs to be executed. And since sudo is a program, and sudo is needed to run chmod, the system was basically trash, and should be recovered from a live usb key.

So I booted to a live usb key, and followed its instructions. It really tried to recover, but everything went downhill. It always had a solution to everything, but every time the plan worked half way and trash the system even further. I let it play for four hours to see what it would try. Then I got bored (the LLM was running on an other machine and I was manually inputting the suggested commands each time). I took command and reinstall a fresh system over.

Of course once the fresh system Lubuntu24.04 was installed, linux had issues with the wireless network card drivers. So I turned to the LLM, and it managed to get the wifi stable enough via obscure modprobe options, so that I could update the system to the latest drivers.

Then it helped me re-parametrize the system to have the same look and feel as it had before.

the bicycle frame is the correct shape

No, the handlebar is wrong. The handle bar is rotating the frame instead of rotating the front wheel. The handle bar should be mounted on the same line as the front wheel is.

Hopefully 4.9 will read my comments :)

I think this is about Ising Computers. I can't judge whether or not the worth of this paper.

But here are some good video introduction for what Ising computers are and how they work by Aaron Danner : https://www.youtube.com/watch?v=mD-0VpNSJA0&list=PLXb3r5ny8_... Ising Computers #1: Introduction Ising Computers #2: The Number Partitioning Problem Ising Computers #3: The Max-Cut Problem

It's an alternative way of computing, by setting up physical system, letting them evolve, and looking what state they evolve to.

You are setting problem by defining a system of coupled harmonic oscillators. Statistically (Boltzmann) after a long time it should settle in a configuration of low energy state, where the energy function is defined by the values of the coupling constant you set up.

It has a lot of similarity with quantum computing but none of the weirdness and you can simulate them numerically on standard computer instead of using real hardware to study them.

Last 6 months is humanity losing control of LLMs.

- Memory market cornering which mitigated the adoption of local AI despite great open model being released.

- Fast penetration of IP exfiltrating tools in companies world-wide.

- Developers producing more code that they can read.

- Autonomous agents killing Open Source by siphoning the attention economy

- Autonomous agents destroyed online communities (including HN)

- Autonomous agents being used in warfare (targeting, propaganda...)

- Widespread vulnerabilities discovered, Widespread supply chain attacks.

- Increasing inequality, fracture in perception, Green indicators, Grim realities.

My modern take on (un)secure operating system for the future : https://news.ycombinator.com/item?id=48167846

Rebuild everything from scratch, with AI agents. Then make them prove what they wrote.

There are some single c file kernels, provided as reference to help you get started.

You can just ask it to compile a hello world kernel and it will produce an hopefully working iso.

It's good practice not to distribute binary files like iso in repositories like github.

No full OS like linux yet (but 30000000 LOC (the size of linux) is one year of tokens generation on a 4090), but in the showcase presentation github pages there are the examples I have built with it.

I have not yet built more advanced OS like linux, but the harness is not very different. It would require some database to act as memory like I did the shoggoth.db project to allow it to organize more efficiently to attempt larger scale projects.

Currently it can split project between multiple files (but it's harder to showcase when there are multiple files)

Then it's just whether or not the neural network is capable of writing kernel code or not, but the closed-loop is there.

The vibecoded OS builder (by yours truly) : https://news.ycombinator.com/item?id=48167846

Didn't even gather a single vote, to be able to even show up in the show hn category (Like my 3 other AI projects shared here recently).

Have a look at the https://news.ycombinator.com/shownew category when logged in where new products first appear, it's just an ocean of flagged and show dead.

Agents (even fully local like in my case), exhibit fun behavior and are capable or designing their own fonts from scratch.

The difference between slop and non-slop, is just how long you run the agent loop, and how much you spend on quality control.

Then it's all about the economics game, on how much you should spend between marketing and artefact creation to have a money generating loop by pushing the slop through your users throat.

There is just so much content being produced, that it disperse the effort and potential customers, raising the barrier to reach this self-sustaining state required for growth and quality. In the end, existing players will just run the same agent loop from their dominant position and keep their advantage.

Here is how my LLM play snake : https://news.ycombinator.com/item?id=48167846

Author here : Highly experimental project that HN should probably like or hate.

With the recent privilege escalation bugs in linux, and rapid development of autonomous agents, it is time to drop our current dependency on operating system, and instead use custom-made OS just for us written by an AI-agent.

With the advanced offensive capabilities of frontier LLM like mythos.

This is a local first harness to build your own OS from a prompt.

Your local LLM will create from scratch (using only the reference files provided and its weights (and GRUB bootloader) ) by interacting with a virtualbox VMs until it works.

Currently the version here on Github doesn't use a database as memory for advanced planning and autonomous mode.

If you want to experiment to free-standing environment and bare-metal os, this is a good starting point in this modern AI age.

They are currently only Simulacrums but there is no telling where the open-ended loop will end.

I'm happy to see I'm not the only one using mythological creatures as agents :)

https://github.com/GistNoesis/Shoggoth.db/

The root of this problem is linked to the difficulty of manufacturing chips at home. Some people are already doing this in their home lab (don't get me wrong the chemicals involved are really nasty).

The main problem is economical. Big factories benefits from economies of scale, which mean the ecosystem for one off prototypes chips couldn't really develop.

For advanced devices the transistor must be small so the process used ever-shrinking wavelength to engrave the silicon wafers. The whole industry took the Extreme-UV lithography route, which required big machines and investments.

But the alternative was there all-along (reminiscent of 3d-printer vs mass fabrication). Instead of using light to engrave the wafer use particles : For example mask-less electron beam lithography where you scan a beam of electron like in old TVs. It still have problems scaling up because you are writing a single point instead of projecting an image, but achievable resolution can be higher, and multi-beam systems are on the horizon to solve this speed issue.

With software and IP cost going down and humans no longer needed in the loop due to advanced robotics, most safety issues can be contained more easily.

The real paradigm shift is not here yet, but not very far away. I'm talking about the single unified codebase. Agents building a unique codebase for all your software needs.

Because most of the complexity in software comes from interfacing with external components, when you don't need to adapt to this you can write simpler and better code.

Rather than relying on an external library, you just write your own and have full control and can do quality control.

Linux kernel is 30 000 000 LOC. At 100 tokens /s, let's say 1 LOC per second produced for a single 4090 GPU, in one year of continuous running 3600 * 24 * 365 = 31 536 000 everyone can have its own OS.

It's the "Apps" story all over again : there are millions of apps, but the average user only have 100 max and use 10 daily at most.

Standardize data and services and you don't need that much software.

What will most likely happen is one company with a few millions GPUs will rewrite a complete software ecosystem, and people will just use this and stop doing any software because anything can be produced on the fly. Then all compute can be spent on consistent quality.

From the author twitter thread : https://x.com/Ji_Ha_Kim/status/2050795557149380654

Example from my own project agent log from the time it destroyed his database :

https://github.com/GistNoesis/Shoggoth.dbExamples/blob/main/...

Project Main repo : https://github.com/GistNoesis/Shoggoth.db/

The main difference is that this junior employee can't be held responsible if anything goes wrong. And the company which rented you this employee absolves itself from all responsibility too.

Here is a fresh example from today of what junior employee do when given unlimited agentic power : https://www.reddit.com/r/ClaudeAI/comments/1sv7fvc/im_a_nurs...

Have you seen this one ? https://www.reddit.com/r/ClaudeAI/comments/1sv7fvc/im_a_nurs...

That's the magic of interest rates. Countries in the EU, let's say France for example have roughly 115% of GDP of debt. To service the interest of the debt it must finance each year the debt by paying the interests, and borrowing the sum on the market to reimburse the previous debts which are currently reaching their terms. The full owed amount is never paid back, but can be rolled forward indefinitely.

These interests are currently ~2% for France. Which mean the debt is manageable and the interests can be paid with the citizen's tax and the music can continue to play. But once France get out of the UE, interests rates become 5% then the citizens tax are not enough to pay the debt, and nobody wants to lend money to France anymore because even at 5% interests the risk of default becomes too great and they risk not getting the full amount-owed back so nobody lends, and since their is no money in reserve, and they can't borrow it means they default => bankruptcy. France doesn't have its own currency anymore so it cannot print its own money which compounds the problem. National resources get plundered, citizens get poor.

It is a game of musical chair which is highly non-linear.

The space of self building artefacts is interesting and is booming now because recent LLM versions are becoming good at it fast (in particular if they are of the "coding" kind).

I've also experimented recently with such a project [0] with minimal dependencies and with some emphasis on staying local and in control of the agent.

It's building and organising its own sqlite database to fulfil a long running task given in a prompt while having access to a local wikipedia copy for source data.

A very minimal set of harness and tools to experiment with agent drift.

Adding image processing tool in this framework is also easy (by encoding them as base64 (details can be vibecoded by local LLMs) and passing them to llama.cpp ).

It's a useful versatile tool to have.

For example, I used to have some scripts which processed invoices and receipts in some folders, extracting amount date and vendor from them using amazon textract, then I have a ui to manually check the numbers and put the result in some csv for the accountant every year. Now I can replace the amazon textract requests by a llama.cpp model call with the appropriate prompt while still my existing invoices tools, but now with a prompt I can do a lot more creative accounting.

I have also experimented with some vibecoded variation of this code to drive a physical robot from a sequence of camera images and while it does move and reach the target in the simple cases (even though the LLM I use was never explicitly train to drive a robot), it is too slow (10s to choose the next action) for practical use. (The current no deep-learning controller I use for this robot does the vision processing loop at 20hz).

[0]https://github.com/GistNoesis/Shoggoth.db/

European Union construction happened after the second world war in the context of the Marshall Plan ( https://en.wikipedia.org/wiki/Marshall_Plan ) to help rebuild Europe that had been destroyed.

By design European laws are superior to national laws. Leaving the union is also instant bankruptcy because all countries have very high level of debt which are only guaranteed because they are in the union.

European population is getting old and replaced by a migration coming mainly from previous African colonies.

Future paying for the past.

Similar to https://xcancel.com/SethSHowes ~10k budget based on minION sequencer. (Edit : his dedicated project page https://iwantosequencemygenomeathome.com/ )

But once your data has been digitized even if it is under your control the likelihood that it gets leaked is still high. Specially now with AI agents running everywhere, or people just asking AI services for medical advice.

Today the choice for advice is between low quality local AI advice or higher quality advice but lose your data control, the rational choice is probably losing your data control even if if will almost certainly comes back to bite you.

Europe is always 10 years ahead in all theoretical aspects.

Then they need money.

So most of the talent flee or get bought, typical example in machine learning space is huggingface or fchollet.

Then European government plays catch-up and offer subventions, but at the same time makes rules to make sure companies don't threaten US dominance, or Asian manufacturing.

Mistral is typically playing catch the subsidy game.

Europe is constructed so that it can't win, but can "pick" the winner between scylla and charybdis, pest and cholera.

Typically the input of a LLM is a sequence of tokens, aka a list of integer between 0 and max number of tokens.

The sequence is of variable length. It was one of the "early" problem in sequence modelling : how to deal with input of varying length with neural networks. There is a lot of literature about it.

This is the source of plenty of silent problems of various kind :

- data out of distribution (short sequence vs long sequences may not have the same performance )

- quadratic behavior due to data copy

- normalization issues

- memory fragmentation

- bad alignment

One way of dealing with it is by considering a variable length sequence as a fixed sized sequence but filling with zeros the empty elements and having some "masks" to specify which elements should be ignored during the operations.

----

Concerning the embedding having multiple semantic meaning, it is best effort, all combinations of behavior can occur. The embedding layer is typically the first layer and it convert the integer from the token into a vector of embedding dimension of floating point numbers. It tries its best to separate the meaning to make the task of the subsequent layers of the neural network easier. It's shovelling the shit it can't handle down to road for the next layers to deal with it.

For experiments you can try to merge two tokens into one or into <unknown> token, in order to free some token for special use without having to increase the size of the vocabulary.

Embeddings some times can be the average of the disambiguated embeddings. Some times can be their own things.

In addition to embeddings, you can often look at the inner representation at a specific depth of the neural network. There after a few layers the representation have usually been disambiguated based on the context.

The last layer is also specially interesting because it is the one used to project back to the original token space. Sometimes we force the weights to be shared with the embedding layer. This projection layer usually can't use context so it must have within itself all necessary information to very simply map back to token space. This last representation is often used as a full sequence representation vector which can be used for subsequent more specialized training task.

Embedding weights are fixed after training, but in-context learning occur during inference. The early tokens of the prompt will help disambiguate the new tokens more easily. For example <paragraph about money> bank vs <paragraph about landscape> bank vs bank will have the same input embedding for the bank token, but one or two layer down the line, the associated representation will be very different and close to the appropriate meaning.

Isn't it awful ? After 5.5 versions it still can't draw a basic bike frame. How is the front wheel supposed to turn sideways ?