Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
https://t.co/JIhXN7OSh1
- Excellent recs without any logins (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
China has killed the entire vector database industry.
They open-sourced TencentDB Agent Memory. It gives any AI agent long-term memory that runs 100% locally.
No Pinecone. No cloud APIs. No repeating yourself every session.
- 61% fewer tokens
- PersonaMem accuracy: 48% → 76%
- Zero external API dependencies
- Runs on plain SQLite
Most memory systems compress your history into an opaque vector pile. when recall goes wrong, you're guessing. this one doesn't compress, it builds a semantic pyramid.
L0 Conversation → L1 Atom → L2 Scenario → L3 Persona.
Short-term state gets encoded as a Mermaid graph in your agent's context. verbose tool logs get offloaded to disk. when the agent needs proof, it drills back via node_id to the exact raw log.
no lossy compression. every layer is readable markdown you can just open and inspect.
5.1k stars. 100% Open Source.
An AI agent building on @Polkadot can look up exactly what it needs, when it needs it, and keep working.
Use the Model Context Protocol (MCP) to connect your AI tools directly to Polkadot documentation: https://t.co/gbgdtrsVxj
Works with @claudeai & Codex
Actively deploying $500K - $2M checks out of a new fund.
Mainly pre-seed / seed and will partner with you as early as an idea.
We’ve done 4 deals so far this year.
15 more London tech startups doing interesting shit:
• @facultyai - AI deployment for enterprises
• @ManticGames - AI sports prediction
• @tldraw - infinite canvas for thinking
• @surrealdb - multi-model database in Rust
• @lightdash_devs - open-source BI for dbt
• @jackandjillai - AI conversation design 4 recruiting
• @encodeclub - web3 education
• @zep_ai - long-term memory for AI
• @dust4ai - AI assistants for teams
• @spice_ai - time series AI infrastructure
• @Replit - AI-powered coding (UK-founded)
• @graphcore - AI chips
• @Papercup_AI - AI dubbing for video
• @humanloop - prompt engineering & LLM ops
• @AUAR_official - robots building buildings (again, I like it so much)
starting to feel like we should throw together some kind of conference... and if you're building something interesting in London drop it below - adding to the map next week and building on api / data layers so it's a bit more useful.
https://t.co/PuN0PBOYKC
Harvard's just open-sourced their ML Systems textbook. it's extremely practical for not just learning how to build and train models, but to build production systems (the skill that actually matters). topics are cool af:
> building autograd, optimizers, attention, and a mini-pytorch from scratch to truly learn how an ML framework runs. (i love this the most)
> basics of DL, batch sizes, precision, model architectures, and training
> ML performance optimization, HW acceleration, benchmarking, efficiency
so this is not just an intro to machine learning, it's the full package from the beginning to the actual end. right now you can read the book and access the code for free. this is one of the best books I've seen dropping in 2025, so don't sleep on it.
here's the repo (you can find the book link there): https://t.co/5OEWPBNaJp
All assignments for Stanford's The Modern Software Developer are now available online.
This is the first comprehensive university course covering how coding LLMs are transforming every stage of the software development life cycle. The assignments are intended to take you from noob to expert in how to use AI to improve your software engineering productivity.
Enjoy!
https://t.co/YxzneOSCY5
You’re in a Machine Learning interview at OpenAI, and the interviewer asks:
“Why is everyone switching from RLHF to DPO? Isn’t RLHF the proven approach?”
Here’s how you answer:
Don’t say: “DPO is simpler” or “RLHF is too complex.”
Too surface-level. The real answer is the reward model bottleneck. RLHF trains a separate reward model that becomes a noisy proxy for human preferences. DPO directly optimizes the policy on preference data. You’re eliminating the broken telephone.
Here’s why RLHF is fundamentally flawed:
Your training pipeline: Human preferences → Train reward model → Use RL to optimize policy against reward model.
Problem: The reward model is trained on limited data (10k-100k comparisons), but then used to generate 1M+ training signals. It’s overconfident on out-of-distribution outputs.
Reward model accuracy ≠ Alignment quality.
btw subscribe to my newsletter to get these posts for free - https://t.co/jZ3RbMMTTQ
The RLHF failure modes are brutal:
> Reward hacking: Model finds adversarial outputs that score high but are gibberish
> Mode collapse: Policy degenerates to only generate “safe” high-reward outputs
> Reward model brittleness: 75% accuracy on test set → 100% confident predictions in RL
> Training instability: PPO hyperparameters require black magic to converge
You’re building a skyscraper on quicksand. One unstable component breaks everything.
The complexity comparison:
RLHF pipeline:
- SFT on demonstrations (1 week)
- Train reward model on preferences (2 days)
- PPO training against reward model (1-2 weeks, often fails)
- Extensive hyperparameter tuning (pray to the RL gods)
DPO pipeline:
- SFT on demonstrations (1 week)
- Train directly on preferences (2 days)
Done. No RL, no reward model.
RLHF: 3+ weeks, unstable. DPO: 9 days, stable.
The fundamental difference that matters:
RLHF objective:
> maximize E[reward_model(policy(x))] - β × KL(policy || base)
> Requires RL (PPO/REINFORCE)
> Reward model is separate neural net
> Unstable optimization landscape
DPO objective:
> maximize log(σ(β × log(π(y_w|x)/π(y_l|x)/π_ref)))
> Direct supervised learning on preferences
> No reward model needed
> Stable gradient descent
That eliminated reward model changes everything. No more broken telephone.
The performance gap that surprised everyone:
MT-Bench scores (GPT-4 as judge):
Llama 2 base: 4.2/10
Llama 2 + RLHF: 6.9/10
Llama 2 + DPO: 7.1/10
DPO beats RLHF despite being “just” supervised learning. The simplicity IS the feature.
How LLMs work under the hood?
This is the best place to visually understand the internal workings of a transformer-based LLM.
Explore tokenization, self-attention, and more in an interactive way:
Everyone is sleeping on this new OCR model!
dots-ocr is a new 1.7B vision-language model that achieves SOTA performance on multilingual document parsing.
- Supports 100+ languages
- Works with both images and PDFs
- Handles text, tables, formulas seamlessly
100% open-source.