Hacker News: xkgt

New comment by xkgt in "Boot a Virtual iPhone via Apple's Virtualization.framework"

xkgt — Sat, 29 Aug 2026 03:36:30 +0000

It is wild that Apple finds it cheaper to build and maintain different App store behaviors than to simplify and make alt app stores available in all markets. Other governments/markets don't care?

State of AI Report 2025

xkgt — Sun, 12 Oct 2025 12:59:12 +0000

Article URL: https://www.stateof.ai/

Comments URL: https://news.ycombinator.com/item?id=45557935

Points: 3

# Comments: 1

New comment by xkgt in "Postgres Just Cracked the Top Fastest Databases for Analytics"

xkgt — Sat, 08 Mar 2025 09:31:55 +0000

I read the title a couple of times and I'm still not sure it isn't misleading. The benchmarks are not just for Postgres but for Postgres with the Mooncake extension. There are also other results for Postgres with different extensions. While it does rank among the top fastest databases, it is not the fastest and not even within the top 10.

DeepSeek's smallpond: Bringing Distributed Computing to DuckDB

xkgt — Tue, 04 Mar 2025 01:09:04 +0000

Article URL: https://mehdio.substack.com/p/duckdb-goes-distributed-deepseeks

Comments URL: https://news.ycombinator.com/item?id=43248947

Points: 198

# Comments: 23

HuggingFace releases support for tool-use and RAG models

xkgt — Wed, 03 Jul 2024 00:47:56 +0000

Article URL: https://github.com/huggingface/transformers/releases/tag/v4.42.0

Comments URL: https://news.ycombinator.com/item?id=40861822

Points: 59

# Comments: 0

New comment by xkgt in "Gemma 2: Improving Open Language Models at a Practical Size [pdf]"

xkgt — Thu, 27 Jun 2024 23:08:07 +0000

Do we know if Gemma models are fundamentally different from the ones hosted as Gemini? Gemini 1.5 flash seems to produce good results for the price and performance.

Intel releases OpenVINO 2024.2 with broader LLM and quantization support

xkgt — Tue, 18 Jun 2024 09:16:43 +0000

Article URL: https://github.com/openvinotoolkit/openvino/releases/tag/2024.2.0

Comments URL: https://news.ycombinator.com/item?id=40715599

Points: 9

# Comments: 0

New comment by xkgt in "Every Way to Get Structured Output from LLMs"

xkgt — Tue, 18 Jun 2024 09:03:43 +0000

I was recently researching structured output generation for my project and I enjoyed using Outlines library a lot. It felt quite fast as it uses FSM and indexing. There are few fine prints though:

1. Sometimes constraints can decrease the quality of the output since syntax of the response is prioritized more than quality of the response 2. For memory constrained inferences, certain sampling strategies like top-k can cause OOM errors if the max_token is too high. I haven't tested that it is entirely due to structured generation but I suppose it is possible for certain regexes. 3. Vision models and other multi-modal models are not supported yet.

Apart from this, closed models also have json output but I am not sure how consistent they are

1. https://platform.openai.com/docs/guides/text-generation/json... 2. https://docs.anthropic.com/en/docs/build-with-claude/tool-us... 3. https://ai.google.dev/gemini-api/docs/api-overview#json

Transformer.js is being embedded in Firefox 130

xkgt — Tue, 04 Jun 2024 23:22:09 +0000

Article URL: https://hacks.mozilla.org/2024/05/experimenting-with-local-alt-text-generation-in-firefox-nightly/

Comments URL: https://news.ycombinator.com/item?id=40579943

Points: 6

# Comments: 1

New comment by xkgt in "Homeschooling (2021)"

xkgt — Mon, 03 Jun 2024 01:50:15 +0000

The author did enroll his kid into school for freshman after a break. Is this normally considered beneficial? Having a period of unschooling followed by traditional school?

New comment by xkgt in "Microsoft Phi-3 Cookbook"

xkgt — Tue, 21 May 2024 23:17:41 +0000

Looks like some of the docs are generated by an llm. I see pictures with typos and imagined terms, incomplete texts etc., I wonder to what extent we can trust rest of the docs.

https://github.com/microsoft/Phi-3CookBook/blob/main/md/04.F...

Geometric Deep Learning – Harnessing Geometric Patterns in Data

xkgt — Sat, 04 May 2024 03:36:51 +0000

Article URL: https://www.youtube.com/watch?v=w6Pw4MOzMuo

Comments URL: https://news.ycombinator.com/item?id=40254597

Points: 4

# Comments: 0

Pats-Wheel: A Passively-Transformable Single-Part Wheel for Robot Navigation

xkgt — Sun, 28 Apr 2024 10:52:47 +0000

Article URL: https://www.youtube.com/watch?v=oZbZZlDDEnc

Comments URL: https://news.ycombinator.com/item?id=40187612

Points: 1

# Comments: 0

Snowflake Launches Text-Embedding Model for Retrieval Use Cases

xkgt — Thu, 25 Apr 2024 23:27:49 +0000

Article URL: https://www.snowflake.com/blog/introducing-snowflake-arctic-embed-snowflakes-state-of-the-art-text-embedding-family-of-models/

Comments URL: https://news.ycombinator.com/item?id=40164309

Points: 24

# Comments: 3

MindEye2: Shared-Subject Models Enable fMRI-to-Image with 1 Hour of Data

xkgt — Tue, 19 Mar 2024 23:09:01 +0000

Article URL: https://arxiv.org/abs/2403.11207

Comments URL: https://news.ycombinator.com/item?id=39761265

Points: 77

# Comments: 30

New comment by xkgt in "The unreasonable effectiveness of i3, or: ten years of a boring desktop env"

xkgt — Sun, 03 Mar 2024 13:42:00 +0000

On a tangential note, I find window management in MacOS much more horrible than Windows. Want to split windows, you end up with full screen. When on multiple monitors, selecting an app on one screen makes the same app active on the other screen (or sometimes it doesn't). I am willing to rewire my habits if I can just figure out how to make Mac window manager behave deterministically. I just don't get what is the grammar of user interaction that the designers went for.

New comment by xkgt in "Microsoft Azure Hit with the Largest Data Breach in Its History"

xkgt — Tue, 20 Feb 2024 02:16:23 +0000

This is really shoddy reporting. The title says Azure data breach whereas the attack is a phishing campaign targeting Office 365. Being a phishing campaign, it is unclear which components of Azure/Microsoft were instrumental in the attack. And the article goes on to make irrelevant allusions to Microsoft's negligent cybersecurity practices. This is such bad reporting that I wonder whether it is done with incompetence or malfeasance.

New comment by xkgt in "Sora: Creating video from text"

xkgt — Fri, 16 Feb 2024 02:43:26 +0000

This is pretty impressive, it seems that OpenAI consistently delivers exceptional work, even when venturing into new domains. But looking into their technical paper, it is evident that they are benefiting from their own body of work done in the past and also the enormous resources available to them.

For instance, the generational leap in video generation capability of SORA may be possible because:

1. Instead of resizing, cropping, or trimming videos to a standard size, Sora trains on data at its native size. This preserves the original aspect ratios and improves composition and framing in the generated videos. This requires massive infrastructure. This is eerily similar to how GPT3 benefited from a blunt approach of throwing massive resources at a problem rather than extensively optimizing the architecture, dataset, or pre-training steps.

2. Sora leverages the re-captioning technique from DALL-E 3 by leveraging GPT to turn short user prompts into longer detailed captions that are sent to the video model. Although it remains unclear whether they employ GPT-4 or another internal model, it stands to reason that they have access to a superior captioning model compared to others.

This is not to say that inertia and resources are the only factors that is differentiating OpenAI, they may have access to much better talent pool but that is hard to gauge from the outside.

New comment by xkgt in "Stable Cascade"

xkgt — Wed, 14 Feb 2024 11:54:04 +0000

This model is built upon the Würstchen architecture. Here is a very good explanation of how this model works by one of its authors.

https://www.youtube.com/watch?v=ogJsCPqgFMk

New comment by xkgt in "Protein biomarkers predict dementia 15 years before diagnosis in new study"

xkgt — Wed, 14 Feb 2024 11:48:34 +0000

> Of 1,463 proteins analysed, aided by with a type of artificial intelligence known as machine learning, 11 proteins were identified and combined as a protein panel, which the researchers have shown to be highly accurate at predicting future dementia.

I understand that press releases are intended for non-technical folks but I don't get the point of this description. Is it assumed that machine learning is less understood than artificial intelligence?