Ripple-65B Preview: U.S. Open-Weight SOTA on Agentic Coding
Commenters note that 'SOTA' is diluted by category narrowing and that the 65B dense model will be slow on consumer hardware.
30 stories
Get a notification every evening
You hear about the digest when it is ready. Turn it off any time.
Notifications are on
You will hear about each evening’s digest when it is ready.
To get notifications on iPhone or iPad, add this site to your Home Screen first: tap Share, then Add to Home Screen. Open it from there to turn notifications on.
Notifications are blocked in this browser. Allow them in site settings and reload.
New notification sign-ups are closed for now. You can follow along with RSS.
Notifications could not be turned on. Check your connection and try again.
Notifications could not be turned off and are still on. Check your connection and try again.
The day’s top discussions
Commenters note that 'SOTA' is diluted by category narrowing and that the 65B dense model will be slow on consumer hardware.
musl's crypt() returns '*' when the key exceeds 256 bytes; '*' is also the field for accounts without password, so login succeeds.
Developers consider long embargo periods dangerous because AI agents rediscover bugs within hours.
A commenter wants more info to understand the distribution, age, and types of vulnerabilities in the codebase.
Commenters appreciate the model team acknowledging the gap between Hy4 and top proprietary models.
A commenter states that all frontier model providers monitor interactions even with activity tracking disabled.
A commenter likes the new API but criticizes the removal of 'MarshalIndent' for making migration difficult.
One commenter suggests the good side of the RAM crisis might be that engineers care about memory usage again.
A commenter points out the crucial part is replacing Gated MLA with DSA to reduce inference cost.
The commenter (author) describes realizing the game state fits in 64k and achieving network sync.
A commenter says they use Claude Code and Opus 5.5 for solo dev work.
A commenter says it's like people are discovering OAuth for the first time.
A commenter says porting a 1993 game won't tell you anything interesting about code structure survival.
A commenter notes benchstat makes a location-shift assumption in its Mann-Whitney U test and doesn't see why it holds.
The project author says Stroq monitors dangerous commands from AI agents; a commenter suggests covering more scenarios with an ebpf backend.
The author introduces Mizu and pymizu packages for lock-free async and parallel computing over shared memory, claiming task round trip drops from 100 µs to under 1 µs.
One commenter suggests non-technical users using LLMs may not be 'cognitive surrender', while another notes free code generation leads to over-engineering.
One commenter warns the automated improvement loop could waste tokens; another asks whether an RL environment or test suite is a better analogy for agent experience.
Commenters note missions in LEO on ISS, research on fluids and seed germination in microgravity, and that semiconductor manufacturing in space is an interesting idea.
Commenters note the simulation doesn't conserve volume and that publishing the method may make it no longer special.
A commenter summarizes Apple's methods to fit an LLM on iPhone: quantization, KV caching, LoRA, sparse MoE model, dedicated AI chip, and offload to cloud.
One commenter claims the initiative is run by a random guy seeking funding, while the author hopes European manufacturers evaluate RVA23 chips.
A commenter suggests we need a term for buying an entire company because it would be cheaper than paying a monthly subscription fee.
One commenter notes that owning the IP space does not prove physical ties to the White House; the network could be a security lab or media guest network.
One commenter suggests adding template support to Kino AI and appreciates that such a product is built by industry experts.
One commenter mentions wrapping generators into an actor system in their own harness and that it worked well.
A smaller, more technical community
A commenter says this is amazing and genuinely fun to play with.
A commenter suspects the slowdown is due to Zig mapping C strings into safe zig arrays.
The week’s most-starred new repositories