justsomeguy1996
Joined 4 years ago
Karma
43
Submissions
9
Member since
May 31
Recent submissions
Comment · 7 months ago
Comment · 7 months ago
I think the main benefits are:
- Slightly improved recall
- Faster index creation
- Online addition of vectors without recalibrating the index
The last point in particular is a big infrastructure win I think.
Comment · 7 months ago
I built a Python implementation of Google's TurboQuant paper (ICLR 2026) for vector search. The key thing that makes this different from PQ and other quantization methods: it's fully data-oblivious. The codebook is derived from math (not trained on your data), so you can add vectors online without ever rebuilding the index. Each vector encodes independently in ~4ms at d=1536.
The repo reproduces the benchmarks from Section 4.4 of the paper — recall@1@k on GloVe (d=200) and OpenAI embeddings (d=1536, d=3072). At 4-bit on d=1536, you get 0.967 recall@1@1 with 8x compression. At 2-bit, 0.862 recall@1@1 with ~16x compression.
Comment · 8 months ago
I built Sourced because LLMs don't reliably know where a package's source code lives. When they need to understand a dependency, they either guess from training data or fall back to web search, which is slow, expensive, and usually just returns user-facing docs — not the actual implementation.
Being able to read the real source code makes a big difference. The model can see how functions actually behave, verify that the APIs it's calling exist and work the way it thinks, and build implementations that are grounded in reality rather than guesswork.
Sourced indexes every package on PyPI (800k+) and npm (3M+), with new releases picked up within 5 minutes. It exposes three tools via MCP — read, grep, and glob — so your agent can navigate source code the same way you would locally. It's lightweight and fast.
Install: curl -sL sourced.dev/install | sh
Works with Claude Code, Cursor, VS Code, Windsurf, Codex, and others.
Comment · 4 years ago
I'm 25 and I studied a computer science BSc. After I graduated I worked as a data scientist for 2 years and now I'm a machine learning engineer at a small start up. I have also been feeling held back from the more senior roles by my lack of mathematics and toyed with the idea of MSc or PhD. I applied to Imperial and UCL Machine Learning MSc on 3 seperate occasions and was rejected from all because of the lack of Maths. In the end it would have cost £16k (course fees for 1 yr) + £26k (loss of earnings for the year after tax) and so I haven't applied since because, although studying those masters programmes would give me a gread mathematical foundation, I can also 100% self-study and save £££.
The habit of studying is the most important habit you can learn at this early stage. If you haven't read it, I strongly advise you read Atomic Habits. It's really helped me. Here are some great mathematics for machine learning courses, some I have taken and some I am currently studying.
Study tips: - Always show up even if you only study for 5 minutes, it's more important to show up than to be perfect - Make studying a daily habit even if you only do it for a short amount of time - Stack studying onto the end of another daily habit, e.g. study after you've eaten breakfast before you start work (this is what I do) - Make studying satisfying (I bought a small calendar and cross of each day I succeeded with a sharpie and it's super satisyfing crossing them off) - Remove distractions by putting your phone on charge in another room - Make notes on new concepts in the form of questions, these can be used later on flash cards to help refresh the material and avoid the effects of the forgetting curve - Write code to see the math in action. Running code is immediately gratifying and makes the value of the maths real and tangible.
- https://www.coursera.org/learn/machine-learning (free, approachable) - https://www.udemy.com/course/machine-learning-data-science-f... (like £10 on sale, very approachable) - https://www.udemy.com/course/mathematics-statistics-of-machi... (like £10 on sale) - https://www.coursera.org/specializations/deep-learning - https://mml-book.github.io/book/mml-book.pdf [textbook, not very approachable until you have some background] (free)
I replicated their methodology on Google's Gemma 4 E4B (open-weight) and released the datasets and code. The repo includes everything needed to build expression probes (detecting expressed emotions) and deflection probes (detecting suppressed emotions), plus an interactive visualiser.
Datasets on HuggingFace: https://huggingface.co/datasets/ryancodrai/emotion-probes