<a href="https:&#x2F;&#x2F;ai.google.dev&#x2F;gemini-api&#x2F;docs&#x2F;models&#x2F;gemini-3.5-flash" rel="nofollow">https:&#x2F;&#x2F;ai.google.dev&#x2F;gemini-api&#x2F;docs&#x2F;models&#x2F;gemini-3.5-flas...</a>

Gemini 3.5 Flash

<a href="https:&#x2F;&#x2F;www.nytimes.com&#x2F;2026&#x2F;05&#x2F;19&#x2F;business&#x2F;google-seach-bar-ai-gemini.html?unlocked_article_code=1.jlA.95yh.ptfBUHf-rBtB&amp;smid=url-share" rel="nofollow">https:&#x2F;&#x2F;www.nytimes.com&#x2F;2026&#x2F;05&#x2F;19&#x2F;business&#x2F;google-seach-bar...</a>, <a href="https:&#x2F;&#x2F;archive.ph&#x2F;XI1sQ" rel="nofollow">https:&#x2F;&#x2F;archive.ph&#x2F;XI1sQ</a><a href="https:&#x2F;&#x2F;techcrunch.com&#x2F;2026&#x2F;05&#x2F;19&#x2F;google-search-as-you-know-it-is-over&#x2F;" rel="nofollow">https:&#x2F;&#x2F;techcrunch.com&#x2F;2026&#x2F;05&#x2F;19&#x2F;google-search-as-you-know-...</a><a href="https:&#x2F;&#x2F;www.theverge.com&#x2F;tech&#x2F;932970&#x2F;google-search-ai-update-io-2026" rel="nofollow">https:&#x2F;&#x2F;www.theverge.com&#x2F;tech&#x2F;932970&#x2F;google-search-ai-update...</a>

Google changes its search box

I’ve built a virtual museum with nearly every operating system you can think of

Mistral AI Acquires Emmi AI to Create the Leading AI Stack

Hi HN, I&#x27;m Antoine Zambelli, AI Director at Texas Instruments.I built Forge, an open-source reliability layer for self-hosted LLM tool-calling.What it does:- Adds domain-and-tool-agnostic guardrails (retry nudges, step enforcement, error recovery, VRAM-aware context management) to local models running on consumer hardware- Takes an 8B model from ~53% to ~99% on multi-step agentic workflows without changing the model - just the system around it- Ships with an eval harness and interactive dashboard so you can reproduce every numberI wanted to run a handful of always-on agentic systems for my portfolio, didn&#x27;t want to pay cloud frontier costs, and immediately hit the compounding math problem on local models. 90% per-step accuracy sounds great, but with a 5-step workflow that&#x27;s a 40% failure rate. No existing framework seemed to address this mechanical reliability issue - they all seemed tailor-made for cloud frontier.Demo video: <a href="https:&#x2F;&#x2F;youtu.be&#x2F;MzRgJoJAXGc" rel="nofollow">https:&#x2F;&#x2F;youtu.be&#x2F;MzRgJoJAXGc</a> (side-by-side: same model, same task, with and without Forge guardrails)The paper (accepted to ACM CAIS &#x27;26, presenting May 26-29 in San Jose) covers the peer-reviewed findings across 97 model&#x2F;backend configurations, 18 scenarios, 50 runs each. Key numbers:- Ministral 8B with Forge: 99.3%. Claude Sonnet with Forge: 100%. The gap between a free local 8B model on a $600 GPU and a frontier API is less than 1 point.- The same 8B local model with Forge (99.3%) outperforms Claude Sonnet without guardrails (87.2%) - an 8B model with framework support beats the best result you can get through frontier API alone.- Error recovery scores 0% for every model tested - local and frontier - without the retry mechanism. Not a capability gap, an architectural absence.I&#x27;m currently using this for my home assistant running on Ministral 14B-Reasoning, and for my locally hosted agentic coding harness (8B managed to contribute to the codebase!).The guardrail stack has five layers, each independently toggleable. The two that carry the most weight (per ablation study with McNemar&#x27;s test): retry nudges (24-49 point drops when disabled) and error recovery (~10 point drops, significant for every model tested). Step enforcement is situational - only fires for models with weaker sequencing discipline. Rescue parsing and context compaction showed no significance in the eval but are retained for production workloads where they activate once in a while.One thing I really didn&#x27;t expect: the serving backend matters. Same Mistral-Nemo 12B weights produce 7% accuracy on llama-server with native function calling and 83% on Llamafile in prompt mode. A 75-point swing from infrastructure alone. I don&#x27;t think anyone&#x27;s published this because standard benchmarks don&#x27;t control for serving backend.Another surprise: there&#x27;s no distinction in current LLM tool-calling between &quot;the tool ran successfully and returned data&quot; and &quot;the tool ran successfully but found nothing.&quot; Both return a value, the orchestrator marks the step complete, and bad data cascades downstream. It&#x27;s the equivalent of HTTP having 200 but no 404. Forge adds this as a new exception class (ToolResolutionError) - the model sees the error and can retry instead of silently passing garbage forward.Biggest technical challenge was context compaction for memory-constrained hardware. Both Ollama and Llamafile silently fall back to CPU when the model exceeds VRAM - no warning, no error, just 10-100x slower inference. Forge queries nvidia-smi at startup and derives a token budget to prevent this.How to try it:- Clone the repo, run the eval harness on a model I haven&#x27;t tested. If you get interesting results I&#x27;ll add them to the dashboard.- Try the proxy server mode - point any OpenAI-compatible client at Forge and it handles guardrails transparently. It&#x27;s the newest model and I&#x27;d love more eyes on it.- Dogfooding led me to optimize model parameters in v0.6.0. The harder eval suite (26 scenarios) is designed to raise the ceiling so no one sits at 100%. Several that did on the original suite can&#x27;t sweep it - including Opus 4.6. Curious if anyone finds scenarios that expose gaps I haven&#x27;t thought of. Paper numbers based on pre v0.6.0 code.Background: prior ML publication in unsupervised learning (83 citations). This paper accepted to ACM CAIS &#x27;26 - presenting May 26-29.Repo: <a href="https:&#x2F;&#x2F;github.com&#x2F;antoinezambelli&#x2F;forge" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;antoinezambelli&#x2F;forge</a>Paper: <a href="https:&#x2F;&#x2F;www.caisconf.org&#x2F;program&#x2F;2026&#x2F;demos&#x2F;forge-agentic-reliability&#x2F;" rel="nofollow">https:&#x2F;&#x2F;www.caisconf.org&#x2F;program&#x2F;2026&#x2F;demos&#x2F;forge-agentic-re...</a> <a href="https:&#x2F;&#x2F;github.com&#x2F;antoinezambelli&#x2F;forge&#x2F;blob&#x2F;main&#x2F;docs&#x2F;forge_ieee_preprint.pdf" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;antoinezambelli&#x2F;forge&#x2F;blob&#x2F;main&#x2F;docs&#x2F;forg...</a>Dashboard: <a href="https:&#x2F;&#x2F;github.com&#x2F;antoinezambelli&#x2F;forge&#x2F;docs&#x2F;results&#x2F;dashboard.html" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;antoinezambelli&#x2F;forge&#x2F;docs&#x2F;results&#x2F;dashbo...</a>

Show HN: Forge – Guardrails take an 8B model from 53% to 99% on agentic tasks

Dumb Ways for an Open Source Project to Die

Apple unveils new accessibility features

Tesla's lithium refinery discharges 231,000 gallons of polluted wastewater a day

Minnesota becomes first state to ban prediction markets

Disney erased FiveThirtyEight

The Setup:<a href="https:&#x2F;&#x2F;i.imgur.com&#x2F;o0hgybh.jpeg" rel="nofollow">https:&#x2F;&#x2F;i.imgur.com&#x2F;o0hgybh.jpeg</a><a href="https:&#x2F;&#x2F;i.imgur.com&#x2F;mcNiomp.jpeg" rel="nofollow">https:&#x2F;&#x2F;i.imgur.com&#x2F;mcNiomp.jpeg</a><a href="https:&#x2F;&#x2F;i.imgur.com&#x2F;vIjw6pc.jpeg" rel="nofollow">https:&#x2F;&#x2F;i.imgur.com&#x2F;vIjw6pc.jpeg</a><a href="https:&#x2F;&#x2F;i.imgur.com&#x2F;nzOwmSC.jpeg" rel="nofollow">https:&#x2F;&#x2F;i.imgur.com&#x2F;nzOwmSC.jpeg</a>

Show HN: Gaussian Splat of a Strawberry

Copy Fail, Dirty Frag, and Fragnesia kernel vulnerabilities

Growing Neural Cellular Automata

Gemini Omni

The Silver Swan

Hey HN! We&#x27;re building Haystack (<a href="https:&#x2F;&#x2F;haystackeditor.com&#x2F;">https:&#x2F;&#x2F;haystackeditor.com&#x2F;</a>) to help teams deal with the explosion in the number of pull requests that need to be reviewed due to the rise of coding agents.Haystack replaces the GitHub PR review system with a queue that triages each PR before a human has to read any diffs. It looks at the diffs, the codebase, and the coding-agent conversation that produced the PR. Haystack then routes it into one of three buckets:1. Safe to merge. This means the PR has enough evidence behind it that the team can merge it without another human&#x27;s review.Some examples:-- A small UI copy change that includes a screenshot showing the final state-- A backend change where the author clearly tested the important paths and ran the changes in a real environment2. Needs fixes. This means that the PR has bugs or violates a rule in your codebase and therefore the PR needs to be fixed by the author.Some examples:-- The agent was asked to make loading a large table faster by adding pagination, but the PR still loads every result at once and &quot;implements&quot; pagination in the UI-- The PR silently catches an error instead of logging, surfacing, or handling it. This violates the team&#x27;s &quot;no silent error swallowing&quot; rule3. Needs human review. This means that the PR could not be sufficiently verified by the author or is touching a sensitive part of the codebase (determined by user-input guidelines) and thus requires human review.Some examples:-- The PR changes a significant amount of logic in billing-- The PR changes an important user flow like onboarding, but the author only ran unit tests and never opened the app to check the flow end-to-end. That violates the team&#x27;s rule that high-impact user-facing changes need manual verification.Instead of starting with line-by-line diffs, Haystack immediately tells the reviewer the goal behind the PR, what design decisions the author made (informed by their coding-agent conversation), and how much the author did to verify that the pull request works (e.g. run scripts, checked the frontend, etc.).In this way, review shifts from &quot;what changed?&quot; to &quot;is this the right behavior and is there evidence that it works?&quot;.Here&#x27;s a quick demo: <a href="https:&#x2F;&#x2F;www.tella.tv&#x2F;video&#x2F;streamlining-code-reviews-with-haystack-65zj" rel="nofollow">https:&#x2F;&#x2F;www.tella.tv&#x2F;video&#x2F;streamlining-code-reviews-with-ha...</a>We previously launched Haystack as a tool for understanding large PRs (<a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=45201703">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=45201703</a>). As many of you can probably relate to, the release of Opus 4.5 completely shattered our conception of how fast an engineer could craft a PR.And as coding agents got even better from 4.5, we realized that pull requests did not scale along with our coding velocity. With each member of our team being able to pump out more than 20 pull requests a day, code review quickly became cognitively exhausting and less helpful.After talking with other folks, we learned many feel similarly, and currently face the binary option of either not doing review at all or trying to keep up with a fire hose of pull requests.Haystack is our attempt at a third path. We still believe in code review, but as coding agents produce more code, human reviewer attention becomes more valuable and more expensive.Haystack helps teams spend that attention on the PRs where a human can meaningfully change the outcome of that PR. And for such PRs, Haystack shows the reviewer what the PR intended to do, whether the author showed that it works, and what design decisions need a second pair of eyes.We&#x27;re still quite early and are figuring out whether Haystack truly makes code review better. We would love any and all feedback!

Show HN: Haystack – Review the PRs that need human attention

The last six months in LLMs in five minutes

Hanoi’s humble beer glass and the memory of a nation

Mini Shai-Hulud Strikes Again: 314 npm Packages Compromised

AI, "Humanity", and Dr. Manhattan Syndrome: A Communications Intervention

KV Sharing, MHC, and Compressed Attention

Show HN: I made a 3D pose maker for artists

Show HN: Pg_deltax, Apache-licensed alternative to TimescaleDB

Polypad

An Apple (II) for Teacher

Crossview 4.4.0 is now available

US transportation bill would add a $130 annual fee for EV drivers

Click (2016)

Predicting categorical&continuous Alzheimer's disease outcomes from 1 MRI scan

Plex's 200% Lifetime Pass price hike tries forcing users to another subscription

9:30:36 PM TUESDAY MAY 19, 2026

THE HACKER NEWS