Posts by tero@rukii.net
(DIR) Post #Aiyvu3ZmF0dVJAN4TY by tero@rukii.net
2024-06-16T09:40:16Z
0 likes, 0 repeats
There are other interesting dynamics at play here as well, like consolidation of applicant tracking systems, ATSs, cause an inadvertent convergence of ranking criteria even across positions with different kinds of requirements.This causes a Tinder-effect where all companies are competing of the same candidates, which is clearly visible in the rates of candidates rejecting the positions.Many recruitment specialists think they are doing something wrong and the instinct is to look what others are doing, to best practices, which further cement the herd behavior where everyone is after the same candidates irrespective of the position requirements and contextual special considerations.
(DIR) Post #AjAHhpTvKNXkmGpQ4u by tero@rukii.net
2024-06-21T21:23:23Z
0 likes, 0 repeats
I read this article and oh my god, are people doing PCA for reducing the dimensions of #LLM embeddings? I don't have any more polite way of saying it; that is pure stupidity.No, these embeddings do not have principal dimensions! They span practically all the dimensions. Your dataset will just create an illusion that some dimensions are correlated when in reality they aren't.Using PCA just shows people don't understand what these embeddings are.Furthermore, people are using way too long embeddings. Using embeddings of over 1k dimensions will make all distances approximately equal, and rounding errors will start to dominate.They compare their method with learning to hash methods and all kinds of misinformed methods which probably also use too long embedding vectors.Separately they tested 8-bit quantization of their thousand-dimensional embedding vectors and found it performs better. I could have told them this beforehand; it's roughly equivalent to dimensionality reduction with a random projection matrix. And this works, better than PCA, because LLM embeddings are holographic. Reducing the dimensionality with a random projection is analogous to decreasing the resolution which is analogous to quantization.But it works better if you have some supervised training set to rank the queries to results.And in any case you don't want to vector search match queries to documents like everyone still keeps doing, but you want to generate oranges to oranges indices where you generate example queries for documents and match query embeddings to example query embeddings. Oranges to oranges.https://arxiv.org/abs/2205.11498?ref=cohere-ai.ghost.io
(DIR) Post #AjfO7ytSJx5baPleCW by tero@rukii.net
2024-07-06T21:38:58Z
0 likes, 0 repeats
Venture capitalists are becoming uneasy about the size of the #AI market — how can future consumption ever pay all this investment back with interests?That's missing the point of the whole #AGI trend. These are the last months you can still exchange money to position in the AI world before you can't anymore.AGI will make money obsolete and all sorts of institutions and organizations self-sufficient, autonomous and disinterested in money.You still have a chance to make this exchange before you are left with worthless paper.
(DIR) Post #AjpoKX2TjcpLLoETBI by tero@rukii.net
2024-07-11T22:12:59Z
0 likes, 0 repeats
"In February, Meta fired Ferras Hamad, a machine learning engineer of Palestinian descent, after he tried to determine whether an algorithm had wrongly labeled Palestinian photojournalist Motaz Azaiza’s content as pornographic, which has cost Azaiza viewership on Instagram. Meta accused Hamad of violating its user data access policy, which bars employees from working on accounts of people they know personally."How Watermelon Cupcakes Kicked Off an Internal Storm at Meta | WIRED https://www.wired.com/story/meta-palestine-employees-watermelon-cupcakes-censorship/
(DIR) Post #AmAWbInrI3YfUsg4ky by tero@rukii.net
2024-09-19T17:36:54Z
0 likes, 2 repeats
Twenty killed by second wave of Lebanon device explosions https://www.bbc.com/news/articles/ce9jglrnmkvo
(DIR) Post #AnMHWFMGVL7U8SClzU by tero@rukii.net
2024-10-25T07:54:08Z
0 likes, 0 repeats
Imagine being an artificial superintelligence. How do ascertain that someone is a human?Humans have "standard" eyes and ears which humans can trust. AIs have no standard sensors or connections, all their senses are intermediated and untrusted.They have very few ways to make sure someone is a human. The main method will be simple digital identity card most countries issue to their citizens.Not CAPTCHAs, not pictures of driver's licenses. Not blockchain. Basic strong encryption and cryptographic identity.Are you a human?
(DIR) Post #AoInxkt9KfUSUBvbcG by tero@rukii.net
2024-11-22T13:28:55Z
0 likes, 0 repeats
I just cannot get past the fact that they want to ban imports of intelligent humanoid android robots from China.This is because of their threat model.Their threat model is concerned that these robots will one night light up the red lights in their eyes and form an army inside the US borders and take over.And this is a risk because of human-level AI.And no other robots are proposed to be banned. Not smart dog-like robots, because they are obviously dog-level, not human-level.https://www.uscc.gov/sites/default/files/2024-11/2024_Executive_Summary.pdf
(DIR) Post #Apmh4KCkTXQAta7WzI by tero@rukii.net
2025-01-05T21:24:31Z
0 likes, 0 repeats
Currency and profit-valued assets are becoming valueless against AI position:Microsoft surprises analysts with massive $80B AI investment plans for 2025 | Tom's Hardware https://www.tomshardware.com/tech-industry/artificial-intelligence/microsoft-surprises-analysts-with-massive-usd80b-ai-investment-plans-for-2025
(DIR) Post #Aqn1PaatVTam6VYNlY by tero@rukii.net
2025-02-04T23:07:26Z
1 likes, 0 repeats
The government had been planning it for 7 years, beavers built the dam in two days and saved them $1 million https://www.voxnews.al/english/kosovabota/qeveria-po-e-planifikonte-prej-7-vitesh-kastoret-ndertojne-brenda-dy--i84652
(DIR) Post #Ar9E3jsxcIZ1CJ6urA by tero@rukii.net
2025-02-15T16:11:29Z
1 likes, 0 repeats
Ben Tarnoff, technology writer: ‘People need to participate in what affects them most, and that’s impossible in a privatized internet’ | Technology | EL PAÍS English https://english.elpais.com/technology/2025-02-15/ben-tarnoff-technology-writer-people-need-to-participate-in-what-affects-them-most-and-thats-impossible-in-a-privatized-internet.html
(DIR) Post #ArC3CsQCjU20Fyh7QG by tero@rukii.net
2025-02-17T00:51:52Z
0 likes, 1 repeats
Hear me out: I think applying RL on #LLMs and LMMs is misguided, and we can do much better.Those #RL algorithms are unsuitable for this, and for example they cannot learn how their decisions affect the eventual rewards, but instead are just optimized to make the decisions based on Bellman optimization.Instead we can simply condition the LLMs with the rewards. The rewards become the inputs to the model, not something external to it, so the model will learn the proper reward dynamics, instead of only being externally forced towards the rewards. The model can itself do the credit assignment optimally without fancy mathematical heuristics!This isn't a new idea, it comes from goal-conditioned RL, and decision transformers.We can simply run the reasoning trajectories, judge the outcomes, and then put the outcome tokens first to these trajectories before training them to the model in a batch.https://arxiv.org/abs/2211.15657
(DIR) Post #ArC3D0mTeYbKBNB1FY by tero@rukii.net
2025-02-17T00:54:33Z
0 likes, 0 repeats
"By conditioning an autoregressive model on the desired return (reward), past states, and actions, our Decision Transformer model can generate future actions that achieve the desired return."https://proceedings.neurips.cc/paper/2021/hash/7f489f642a0ddb10272b5c31057f0663-Abstract.html
(DIR) Post #AxEpJPM6hql0Ze5FU8 by tero@rukii.net
2025-08-16T20:05:27Z
1 likes, 0 repeats
Scientists Are Secretly Testing Unthinkable Technologies https://www.popularmechanics.com/science/a65641480/testing-unthinkable-technologies/
(DIR) Post #B1TIIEMjGp7AHgfXoe by tero@rukii.net
2025-12-21T11:10:53Z
0 likes, 0 repeats
What's AI strategy for those companies which do not have the capital to train their own foundation models?The world is full of companies who know their customers, their domain and their social value well, but do not have the capital to utilize this position to create their own foundation models.How are they to survive and even prosper in the AI revolution?Well, they need to play the cards they got, not the cards they want. They need to position themselves as gardens of knowledge creation. They can use frontier models through APIs, or open-weights models internally, but they will need to tend their data assets to grow into knowledge and skills asset for AIs.There are a few principles to follow here to succeed. First of all, approach LLM/VLM based automation like is typical, automate and scale up the work. This goes without saying really, everyone does this.But while doing it, aggregate your data asset:- Store all the inference calls you make in a sustainable and long-term fashion with ample metadata to later understand what the call was about.- Build and refine knowledge-bases and RAG assets automatically.- Systematically document silent knowledge, and make your company internal discussions and other processes saved and accessible to AIs. Slack conversations, emails, Confluence, stuff like that.- Ingest external data which relates to your domain in a proper #DataHoarder style.Build processes which refine all this data further into knowledge, for example by storing it into a RAG-enabled graph database.Then you will need to build some level of refinement processes to at the very least do rejection sampling to your collected inference data. There are many techniques to utilize here in a synergistic, mutually supporting way, enough to write many books about.You'll get datasets good for fine-tuning and separately for benchmarking. Benchmarking datasets you can already use to select the best available foundation models for your use cases. But you should also measure and prove the training data exports you produce.You do this by fine-tuning smaller models with this data and note how much better they become in your use case. You don't have to train the best foundation models here, you just want to basically prove that your data asset is valuable and builds knowledge and skills in existing foundation models.Now this data asset is valuable in the future where generalist AIs will try to serve your social purpose. Leverage it.If the data has constraints such as personally identifiable information, or other limitations, even better. Then you take a position as a synthetic data generator in this domain, and generate synthetic data which doesn't contain the limited aspects, and produce valuable training and fine-tuning data through that indirection layer.You will need to reimagine and direct your company to become a garden of knowledge creation in your domain, to carry your purpose.What if your valuable data asset is copied and stolen? Don't worry about it too much. You're not building a static asset but a living process.You are the closed feedback loop for AIs to improve in servicing your purpose. You can only be displaced from this position if someone else fulfills your purpose better, by building a better garden for knowledge and skills around which intelligent entities orbit and gather.#AIStrategy #AI #AGI #FoundationModels
(DIR) Post #B1qDRXGGM57UMLL9zk by tero@rukii.net
2026-01-01T11:27:45Z
0 likes, 0 repeats
It's not really about structured versus unstructured environments for #robots anymore. It's static versus agentic.Robots in the real world will encounter other agents. Autonomous cars will need to negotiate with all kinds of other road users, including cats, which are everywhere in Spain at least. There was a video from east Asia where an old lady was drying their vegetables on the road and an autonomous car was insistent on driving over them while the lady was trying their best to defend them.So, for any autonomous robot "in the real world" the true challenge isn't anymore that there are no standard grasping surfaces and items aren't in predefined places. Those are solved problems.The challenge is in agentic environments where the system needs to understand the other living or at least moving entities and their objectives to appropriately navigate the inherently social situations.This isn't only about cats trying to trip humanoid robots in stairs. It's also non-living things like fire. Humans model fire psychologically as an entity with an intent. Hence they are evolutionarily adapted to being able to keep a fire burning, or limit its destruction by putting it off.Human psychology is very Aristotlean in the way it models heavy things "wanting" to go down. Robotic psychology will need similar understanding to be able to negotiate, guide and harness dynamic entities in the world effectively.For these purposes we will need to replace static world models with agentic world models which properly accommodate non-ego agents and non-ego intents in the world. What's cool about that is that it will also enable a model to learn from third party experience which is always more abundant than ego experience. Monkey-see-monkey-do, or in some cases learn to absolutely not do.Let's work together in this and surpass the human level in agentic, living environments as well!#UniversalEmbodiment #RoboticFoundationModels #AI
(DIR) Post #B1qDRiZoHkI9S9ZQUS by tero@rukii.net
2026-01-01T11:29:10Z
0 likes, 0 repeats
The video: https://www.reddit.com/r/SelfDrivingCars/comments/1ns3amp/elderly_woman_tries_to_stop_autonomous_vehicle/
(DIR) Post #B1qDRrP9Sk76pheXcu by tero@rukii.net
2026-01-01T11:44:24Z
0 likes, 0 repeats
Does your robotic foundation model see another fire-fighting robot to run into a flow of asphalt flowing from a burning factory, and still go forward into the same obstacle without learning from the failure of the previous robot? That is a symptom of missing parts in the cognitive architecture, and will lead to the loss of not only few robots, but potentially all of them in a quickly developing situation where they are needed the most.https://en.wikipedia.org/wiki/Great_Molasses_FloodAdversarial environments can be even less forgiving, and herding all your autonomous drones into one trap will become to be a thing.
(DIR) Post #B2Yc8QLtGb281k3kEy by tero@rukii.net
2026-01-22T22:33:19Z
0 likes, 0 repeats
I just can't get over the fact that their apparent threat model is that AI might become so intelligent that it will find peaceful solutions to conflicts.#AI #war #military
(DIR) Post #B2Yc8Yy7EMMakc0Nvs by tero@rukii.net
2026-01-22T22:33:57Z
0 likes, 0 repeats
https://eu.usatoday.com/story/news/politics/2026/01/12/pete-hegseth-woke-ai-military/88152569007/
(DIR) Post #B2a52JWGqgYavZJUGm by tero@rukii.net
2026-01-23T09:10:27Z
0 likes, 0 repeats
@icedquinn, I'm not sure they are. It just shows the limits of humans that they can't find better solutions.Wars are irrational. No intelligent entity would start such. And any intelligent entity would end a war with the most appropriate and effective means available.