Feed aggregator

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

Hacker News - Fri, 03/06/2026 - 12:03am

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.

Comments URL: https://news.ycombinator.com/item?id=47271099

Points: 2

# Comments: 0

Categories: Hacker News

Show HN: WingNews – Htmx Hacker News Reader

Hacker News - Thu, 03/05/2026 - 11:51pm

WingNews is a dark mode Hacker News reader client built with HTMX and Go. Any suggestions are greatly appreciated.

Comments URL: https://news.ycombinator.com/item?id=47271019

Points: 1

# Comments: 0

Categories: Hacker News

Ask HN: Wish Linux tmpfs support compression option

Hacker News - Thu, 03/05/2026 - 11:51pm

I run lots of unprivileged containers, and apps inside create tons of temporary files. tmpfs is perfect for this (easy to mount in unprivileged containers). Adding a compression feature would help a lot.

zram needs root to manage. mkfs.btrfs setup feels way too heavy for what is basically “compressed /tmp”.

Why has tmpfs never gotten an official compression feature?

Comments URL: https://news.ycombinator.com/item?id=47271012

Points: 1

# Comments: 0

Categories: Hacker News

Create PDF Resume

Hacker News - Thu, 03/05/2026 - 11:43pm

Article URL: https://createpdfresume.com/

Comments URL: https://news.ycombinator.com/item?id=47270955

Points: 1

# Comments: 0

Categories: Hacker News

Show HN: Yappy – A Python TUI to automate LinkedIn yapping

Hacker News - Thu, 03/05/2026 - 11:22pm

Hey HN,

I got tired of the performative culture on LinkedIn. The platform mostly feels like people just yapping at each other all day to farm engagement, so I decided to build a CLI tool to do the yapping for me. It's called Yappy (because we just yap yap yap).

It is an open-source Python TUI that automates your LinkedIn engagement directly from the command line. It logs in, reads your feed, and uses the Gemini API to generate context-aware comments and drop likes based on your prompt parameters. Everything runs completely in the terminal, so you never actually have to look at the feed yourself.

I know AI-generated comments and web scraping are controversial, and LinkedIn's TOS strictly prohibits this kind of automation. This is built purely as an experimental toy project and a proof-of-concept for integrating LLMs with terminal interfaces. If you actually decide to run it, definitely use a burner account, as LinkedIn will likely restrict you if you run it too aggressively.

I am mostly looking for technical feedback on the Python code architecture and the TUI implementation.

Would love to hear your thoughts. Roast my code or drop a PR if you find the concept funny.

Comments URL: https://news.ycombinator.com/item?id=47270845

Points: 2

# Comments: 0

Categories: Hacker News

Shut Up and Take My Money

Hacker News - Thu, 03/05/2026 - 11:17pm
Categories: Hacker News

Spell UI

Hacker News - Thu, 03/05/2026 - 11:16pm

Article URL: https://spell.sh/

Comments URL: https://news.ycombinator.com/item?id=47270816

Points: 2

# Comments: 0

Categories: Hacker News

Show HN: Swarm – Program a colony of 200 ants using a custom assembly language

Hacker News - Thu, 03/05/2026 - 11:15pm

We built an ant colony simulation as an internal hiring challenge at Moment and decided to open it up publicly.

You write a program in a custom assembly-like (we call it ant-ssembly) instruction set that controls 200 ants. Each ant can sense nearby cells (food, pheromones, home, other ants) but has no global view. The only coordination mechanism is pheromone trails, which ants can emit and sense them, but that's it. Your program runs identically on every ant.

The goal is to collect the highest percentage of food across a set of maps. Different map layouts (clustered food, scattered, obstacles) reward very different strategies. The leaderboard is live.

Grand prize is a trip to Maui for two paid for by Moment. Challenge closes March 12.

Curious what strategies people discover. We've seen some surprisingly clever emergent behavior internally.

Comments URL: https://news.ycombinator.com/item?id=47270807

Points: 9

# Comments: 3

Categories: Hacker News

Show HN: SafeAgent – exactly-once execution guard for AI agents

Hacker News - Thu, 03/05/2026 - 11:14pm

LLM agents and tool-calling systems frequently retry or replay actions.

This becomes dangerous when the action is irreversible: - sending emails - opening tickets - executing trades - triggering payouts

SafeAgent is a small Python library that enforces exactly-once execution for agent actions using request-ID deduplication.

Example:

from settlement.settlement_requests import SettlementRequestRegistry

registry = SettlementRequestRegistry()

def send_email(): print("EMAIL SENT")

registry.execute( request_id="email:user123:invoice", action="send_email", payload={"to": "user123@example.com"}, execute_fn=send_email )

If the same request_id is replayed, the action does not run again.

Instead the original execution receipt is returned.

pip install safeagent-exec-guard

GitHub: https://github.com/azender1/SafeAgent

PyPI: https://pypi.org/project/safeagent-exec-guard/

Comments URL: https://news.ycombinator.com/item?id=47270797

Points: 2

# Comments: 0

Categories: Hacker News

Pages