frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Gemini-flash-latest silently broke Search grounding for 1 month

2•takibboinz•2h ago
On January 21, Google quietly changed the gemini-flash-latest alias to point to gemini-3-flash-preview — a model that does not support Google Search grounding.

The API never returned an error. HTTP 200, valid JSON, finish reason: STOP. The only thing missing was groundingMetadata. No warning, no deprecation notice, nothing.

I run PartsplanAI, a B2B electronic components marketplace. Grounding is not optional for us — we use it to verify part specs against real datasheets and prevent hallucination. Wrong capacitance values or voltage ratings from a language model aren't just embarrassing; they cause real problems for engineers downstream.

For approximately one month (late January through February 27), our AI features ran without any grounding. Every part recommendation, every spec search — generated purely from the model's pre-trained knowledge. The corrupted data accumulated in our database and was served to B2B customers. We had no idea.

On February 27, I noticed recommendations weren't matching real datasheets. What followed was 16 hours of debugging — 63 Git commits, 13 different approaches. I rewrote prompts, rebuilt the search pipeline, changed configurations, adjusted timeouts, switched between parallel and sequential calls. Nothing worked, because the problem was never in my code.

The fix: switch to gemini-2.5-flash. 20 minutes. Done.

The changelog entry for January 21 reads only: "gemini-flash-latest alias now points to gemini-3-flash-preview"

No mention of grounding regression. No compatibility warning.

There's also a second undocumented behavior: on gemini-2.5-flash, if you set responseMimeType: 'application/json' and googleSearch simultaneously, the search is silently ignored. No error, no docs, no warning.

GitHub Issue #384 (google-gemini/generative-ai-js) confirms the grounding issue was known in the community before the alias change was made.

The January 21 changelog was published the same week as the Gemini 2.0 Flash general availability announcement.

If you're using gemini-flash-latest with grounding, verify that groundingMetadata is actually present in your responses. You may have been affected since January 21.

The plan to kill Ali Khamenei

https://www.ft.com/content/bf998c69-ab46-4fa3-aae4-8f18f7387836
1•nsoonhui•48s ago•0 comments

Show HN: Free in-browser sprite sheet packer and slicer with JSON export

https://jaconir.online/tools/sprite-sheet-generator
1•dafin•1m ago•0 comments

Knuth Test using Claude Sonnet 4.6 problem 1.1.3

2•daly•3m ago•0 comments

How are teams managing Windows devices remotely today?

1•elinacorval•3m ago•0 comments

SleepSort O(n)

https://old.reddit.com/r/programminghorror/comments/lgsd18/i_present_sleepsort/
1•ggeorgovassilis•6m ago•0 comments

Mac Studio 512GB RAM Option Disappears Amid Global DRAM Shortage

https://www.macrumors.com/2026/03/05/mac-studio-no-512gb-ram-upgrade/
1•tosh•7m ago•0 comments

Attempted suicides, fights, pain: 911 calls reveal misery at ICE facility

https://apnews.com/article/suicide-ice-detention-centers-b2d1cb0e4b579e0d89caabd00aa04e34
1•petethomas•11m ago•0 comments

Hit and Run

https://randsinrepose.com/archives/hit-and-run/
1•srijan4•11m ago•0 comments

Show HN: Reelforge – AI tool for generating TikTok and Reels ad scripts

https://reelforge-ai1.vercel.app/
1•Milkoo•11m ago•0 comments

How to Scrape NPI Healthcare Provider Data Without an API Key

1•vhubsystems•12m ago•0 comments

Show HN: Plexicus now has a free tier

https://www.plexicus.ai/
1•anwarfirst•12m ago•0 comments

Show HN: VocabPop – Simple Word Game

https://www.vocabpop.com/
1•mb69•13m ago•0 comments

Knuth Test Using Claude Sonnet 4.6 Problem 1.1.2

1•daly•13m ago•0 comments

I've been thinking a bit about continual learning

https://twitter.com/awnihannun/status/2029672507448643706
1•tosh•15m ago•0 comments

Show HN: CardRooms, a Competitive Cardgame Platform

https://www.cardrooms.app/
1•kinskrig•16m ago•0 comments

Show HN: Openboard – Live public metrics page for open startups

https://openboard.fyi/
1•ardakaan•17m ago•0 comments

Uber and Walmart customer data at risk as its vendor Woflow gets compromised

https://securityboulevard.com/2026/03/shinyhunters-claims-woflow-breach-what-it-means-for-saas-su...
1•rudderdev•18m ago•0 comments

Show HN: EV range calculator – range circles and charger map (700 models)

1•sthorat•19m ago•0 comments

Robot 'mice' developed to inspect hadron collider

https://www.bbc.com/news/articles/cddnn261n60o
2•dabinat•20m ago•0 comments

Typst Examples Book

https://sitandr.github.io/typst-examples-book/book/
1•slow_typist•22m ago•0 comments

Knuth Test Using Claude Sonnet 4.6 problem 1.1.1

1•daly•23m ago•0 comments

Show HN: A free personality test based on five elements

https://fiveelementstest.com
1•YarkYao•24m ago•0 comments

Show HN: 1v1 coding game that LLMs struggle with

https://yare.io
2•levmiseri•25m ago•0 comments

P-Stream Shutdown

https://rentry.co/rkkvxesm
1•genderdoog•25m ago•0 comments

Netflix AI Filmmaking on a Spree

https://entrepreneurloop.com/netflix-ai-filmmaking-interpositive-ben-affleck-acquisition/
1•rahuldogra•25m ago•0 comments

AI Tooling for Software Engineers in 2026

https://newsletter.pragmaticengineer.com/p/ai-tooling-2026
1•ColinEberhardt•27m ago•0 comments

Humans and Agents in Software Engineering Loops

https://martinfowler.com/articles/exploring-gen-ai/humans-and-agents.html
1•ColinEberhardt•29m ago•0 comments

Crypto bill hits new impasse, raising doubts over its future

https://www.reuters.com/business/finance/crypto-bill-hits-new-impasse-raising-doubts-over-its-fut...
1•petethomas•29m ago•0 comments

Zammad open-source helpdesk introduces AI without LLM lock-in

https://zammad.com/en/company/press/zammad-introduces-ai-features-with-free-choice-of-large-langu...
1•JulieDenie•29m ago•0 comments

Fixing a major evaluation order footgun in Rye 0.2

https://ryelang.org/blog/posts/rye-0.2-whats-new/
2•todsacerdoti•30m ago•0 comments