ggnore7452

Yes! Really nice to see people here at HN catch the Starship Troopers homage. These days many people's first reaction is Helldivers, which of course got that same tone from Starship Troopers too. And yes, the trailer was made with AI video tools.

Thank you so much for sharing it, really made my day!

Ran a few quick tests on my web-fetch bench. As much as I like DeepSeek, its search and fetch aren't very good (is quite cheap tho). You'd probably be better off using free services like Jina directly, or if you're going paid, Parallel, Firecrawl, Tavily and Exa are all pretty good now.

Isn't that the memory files are loading by default?

too bad that only the smaller on-device models support native audio input.

imo, for health related stuff. or most of the general knowledge doesn't require latest info after 2023. the internal knowledge of LLM is so much better than the web search augmented one.

I’m fine with hires without degrees. But if Google still filters people with LeetCode style coding questions, what’s the point of that in this day and age?

so immersive i actually hit ctrl+w and closed the whole tab.

I’ve done a similar PDF → Markdown workflow.

For each page:

- Extract text as usual.

- Capture the whole page as an image (~200 DPI).

- Optionally extract images/graphs within the page and include them in the same LLM call.

- Optionally add a bit of context from neighboring pages.

Then wrap everything with a clear prompt (structured output + how you want graphs handled), and you’re set.

At this point, models like GPT-5-nano/mini or Gemini 2.5 Flash are cheap and strong enough to make this practical.

Yeah, it’s a bit like using a rocket launcher on a mosquito, but this is actually very easy to implement and quite flexible and powerfuL. works across almost any format, Markdown is both AI and human friendly, and surprisingly maintainable.

I’ve been using LLMs for this kind of geo-guessing since Gemini 2.0. Even without access to internet search like o3, they perform surprisingly well.

appreciate the question on hparams for websearch!

one of the main reasons i build these ai search tools from scratch is that i can fully control the depth and breadth (and also customize loader to whatever data/sites). and currently the web search isn't very transparent on what sites they do not have full text or just use snippets.

having computer use + websearch is definitely something very powerful (openai's deep research essentially)

How’s this compare to likes of Fish audio? Wish they support voice clone using longer audio tho .

Haven’t looked into this space for few months , but iirc, previously SOTA was like GPT VITS or something ?

anyone tried this? is this actually overall better than xgboost/catboost?

if anything i would consider embeddings bit overrated, or it is safer to underrate them.

They're not the silver bullet many initially hoped for, they're not a complete replacement for simpler methods like BM25. They only have very limited "semantic understanding" (and as people throw increasingly large chunks into embedding models, the meanings can get even fuzzier)

Overly high expectations lets people believe that embeddings will retrieve exactly what they mean, and With larger top-k values and LLMs that are exceptionally good at rationalizing responses, it can be difficult to notice mismatches unless you examine the results closely.