Alternatives
Products that do what Learning to (Learn at Test Time) does
I’m excited to share a project I’ve been working on for over a year, which I believe will fundamentally change our approach to language models. We’ve designed a new architecture, which replaces the hidden state of an RNN with a machine learning model. This model compresses context through actual gradient descent on input tokens. We call our method “Test-Time-Training layers.” TTT layers directly replace attention, and unlock linear complexity architectures with expressive memory, allowing us to train LLMs with millions (someday billions) of tokens in context. Our instantiations, TTT-Linear…
- 1IB
Built a ~9M param LLM from scratch to understand how they actually work. Vanilla transformer, 60K synthetic conversations, ~130 lines of PyTorch. Trains in 5 min on a free Colab T4. The fish thinks the meaning of life is food. Fork it and swap the personality for your own character.
Apr 2026 · github.com
- 2
First TTS model to support all 22 Indic languages + English
2024
- 3D3
I replicated David Ng's RYS method (https://dnhkng.github.io/posts/rys/) on consumer AMD GPUs (RX 7900 XT + RX 6950 XT) and found something I didn't expect. Transformers appear to have discrete "reasoning circuits" — contiguous blocks of 3-4 layers that act as indivisible cognitive units. Duplicate the right block and the model runs its reasoning pipeline twice. No weights change. No training. The model just thinks longer. The results on standard benchmarks (lm-evaluation-harness, n=50): Devstral-24B, layers 12-14 duplicated once: - BBH Logical Deduction: 0.22 → 0.76…
Mar 2026 · github.com
- 4
- 5IB
I built a voice agent from scratch that averages ~400ms end-to-end latency (phone stop → first syllable). That’s with full STT → LLM → TTS in the loop, clean barge-ins, and no precomputed responses. What moved the needle: Voice is a turn-taking problem, not a transcription problem. VAD alone fails; you need semantic end-of-turn detection. The system reduces to one loop: speaking vs listening. The two transitions - cancel instantly on barge-in, respond instantly on end-of-turn - define the experience. STT → LLM → TTS must stream. Sequential pipelines are dead on arrival for natural…
Mar 2026 · ntik.me
- 6AB
I built AutoThink, a technique that makes local LLMs reason more efficiently by adaptively allocating computational resources based on query complexity. The core idea: instead of giving every query the same "thinking time," classify queries as HIGH or LOW complexity and allocate thinking tokens accordingly. Complex reasoning gets 70-90% of tokens, simple queries get 20-40%. I also implemented steering vectors derived from Pivotal Token Search (originally from Microsoft's Phi-4 paper) that guide the model's reasoning patterns during generation. These vectors encourage behaviors like numerical…
2025
- 7
- 8CT
2018 · github.com
- 9
- 10BG
Trying to do gradient descent using automatic differentiation over branchy programs? Or to combine them with neural networks for end-to-end training? Then this might be interesting to you. We develped DiscoGrad, a tool for automatic differentiation through C++ programs involving input-dependent control flow (e.g., "if (f(x) < c) { ... }", differentiating wrt. x) and randomness. Our initial motivation was to enable the use of gradient descent with simulations, which often rely heavily on such discrete branching. The latter makes plain autodiff mostly useless, since it can only account for the…
2024 · github.com
- 11IR
Hey HN! I built a proof-of-concept for AI memory using Git instead of vector databases. The insight: Git already solved versioned document management. Why are we building complex vector stores when we could just use markdown files with Git's built-in diff/blame/history? How it works: Memories stored as markdown files in a Git repo Each conversation = one commit git diff shows how understanding evolves over time BM25 for search (no embeddings needed) LLMs generate search queries from conversation context Example: Ask "how has my project evolved?" and it uses git diff to show actual…
2025 · github.com
- 12FC
Hi HN! I've found this visualization tool immensely helpful over the years for getting an intuition for how an LLM "sees" some piece of text, and with a bit of elbow grease decided to move all compute to client side so I could make it publicly available. I've found it particularly useful for - Understanding exactly how repetition and patterns affect a small LM's ability to predict correctly - Understanding different tokenization patterns and how it affects model output - Getting a general sense of how "hard" different prediction tasks are for GPT-style models Known problems (that I probably…
2023 · perplexity.vercel.app
- 13IB
We wanted to do something very challenging to prove to ourselves that we can do anything we put our mind to. The reasoning for why we chose to build a toy TPU specifically is fairly simple: - Building a chip for ML workloads seemed cool - There was no well-documented open source repo for an ML accelerator that performed both inference and training None of us have real professional experience in hardware design, which, in a way, made the TPU even more appealing since we weren't able to estimate exactly how difficult it would be. As we worked on the initial stages of this project, we…
2025 · tinytpu.com
- 14AS
We explored a novel method to gauge the significance of tokens in prompts given to large language models, without needing direct model access. Essentially, we just did an ablation study on the prompt using cosine similarity of the embeddings as the measure. We got surprisingly promising results when comparing this really simple approach to integrated gradients. Curious to hear thoughts from the community!
2023 · heatmap.demos.watchful.io
- 15FD
I worked on this applied Deep Reinforcement Learning course for the better part of 2021. I made a Datacamp course [0] before, and this served as my inspiration to make an applied Deep RL series. Normally, Deep RL courses teach a lot of mathematically involved theory. You get the practical applications near the end (if at all). I have tried to turn that on its head. In the top-down approach, you learn practical skills first, then go deeper later. This is much more fun. This course (the first in a planned multi-part series) shows how to use the Deep Reinforcement Learning framework RLlib to…
2022 · courses.dibya.online
- 16
- 17HW
2011 · quizlet.com
- 18NT
What inspired this project today was watching this amazing video by 3Blue1Brown called "But what is a GPT?" on Youtube (https://www.youtube.com/watch?v=wjZofJX0v4M - I highly recommend watching it). I added it to the repo for reference. When it clicked in my head that "knowing a fact" is nearly synonymous with predicting a word (or series of words), I wanted to put it to the test, because it seemed so simple. I chose JavaScript because I can exploit the way it structures objects to aid in the modeling of language. For example: "I want to be at the beach", "I will do it later",…
2024 · github.com
- 19FG
We developed a new framework that enables flexible control of generated text in language models. By combining several models and/or system prompts in one mathematical formula, it lets you tweak your style and combine model outputs with ease. A handy tool for those working with LLMs, looking for more fine-grained control of stylistic output. More details in our paper: https://arxiv.org/abs/2311.14479. Feedback and potential applications are welcome.
2023 · github.com
- 20IB
Hello Hacker News, When learning foreign languages, I made the most progress by speaking them throughout the day, every day. So I made a site where you can *speak* to an AI language teacher to practice both listening and speaking. # The product *What I have now:* * Multilingual speech recognition: You can ask a question in English and get an answer in your target language. * Feedback on your grammar. * Suggestions: See examples of what to say next to keep the conversation flowing. * Speed: Choose a lower speed for beginners or a faster one for advanced levels. * Translations: Click to see a…
2023 · gliglish.com
- 21PE
Spelltest framework simulates conversations between AI ‘synthetic users' in an environment to test and refine LLM-based applications. It ensures your app converse with utmost accuracy and relevance. Post-chat, Spelltest assesses responses, providing qualitative and quantitative feedback on performance. Suitable for both chat and completion modes. When to use: - After modifying your prompt. - When your LLM provider updates. - As a CI step for you repo. All feedback and collaborations appreciated!
2023 · github.com
- 22IW
2024 · gitlab.com
- 23PG
I’m Andrew, co-founder of Recall. Over the past few days I’ve been building Predict, a playground where anyone can: - propose skills we should measure in language models—live examples include difficult math, memory-manipulation resistance, code generation, and empathy under bad news - write evals (graded prompts) for those skills - forecast which models will score highest once GPT-5 is released Why this exists Benchmarks leak into training data quickly; scores are unreliable and labs still declare progress. The prediction tool aims keeps the target moving by letting the crowd define both the…
2025
- 24ML
Aug 2026 · github.com
Ranked by how close each launch is in meaning, then by votes. Refine with a description →