HNHacker News
TopNewBestAskShowJobs

qrios

227 karma · joined January 19, 2013

It is not a question if, but ‚when‘ we will see what we can call AGI. The ‚when‘ can be answered if we know the ‚how‘. The ‚how‘ we know if the ‚when‘ is answered by ‚today‘.
submissionscomments
qrios··on SmolGPT: A minimal PyTorch implementation for training a small LLM from scratch
It is downloading a dataset (TinyStories) from huggingface[1]. Here you can drill down deeper into the structure and content of the source data.

[1] https://huggingface.co/datasets/roneneldan/TinyStories

qrios··on SmolGPT: A minimal PyTorch implementation for training a small LLM from scratch
Looks great!

Easy to put into a container and the first training with the defined data is already running (on a 16core, 128GB, Nvidia 3090 with 14GB available).

qrios··on First AI Benchmark Solved Before Release: The Zero Barrier Has Been Crossed
Since model capabilities do not change after release, shouldn't the model release date be the benchmark? (In this case, o1-preview was released on September 12, 2024)
qrios··on I Stopped Using Kubernetes. Our DevOps Team Is Happier Than Ever
https://news.ycombinator.com/item?id=42234097
qrios··on A stubborn computer scientist accidentally launched the deep learning boom
> Huang argued that the simple existence of CUDA would enlarge the supercomputing sector. This view was not widely held, and by the end of 2008, Nvidia’s stock price had declined by seventy percent…

This is a good example of how investor behaviour can only quantitatively project what will happen in the future. Huang's bet on GPUs for high performance computing made sense in the long term.

Intel didn't have the staying power with the i860[1] a decade earlier (and of course had no idea how to offer decent developer tools). I tried really hard to develop meaningful and executable programmes with an 8bit card (DSM). CUDA was a revelation for me.

[1] https://www.geekdot.com/intel-80860/

qrios··on iPhones Mysteriously Rebooting Themselves, Locking Cops Out
https://news.ycombinator.com/item?id=42081874

Edit: wrong link

qrios··on My Number One Issue: The anti-intellectual turn in higher education
At least for Europe it is applicable too. The Bologna process aligned the whole landscape with the very basics of the US system.
qrios··on Ollama 0.4 is released with support for Meta's Llama 3.2 Vision models locally
> Specifically server.cpp has been excised (which was deprecated by llama.cpp anyway)

Is there any more specific info available about who (llama.cpp or Ollama) removed what, where? As far as I can see, the server is still part of llama.cpp.

And more generally: Is this the moment when Ollama and Llama part ways?

qrios··on Useful built-in macOS command-line utilities
The point here is to open a document with an app not assigned as default for the given mime-type by file name extension.
qrios··on Sugar Rationing Lowered Diabetes and Hypertension in British Children
https://archive.is/CJfyO
qrios··on Before you can have Smalltalk, you must first defeat capitalism?
> Some people, unbound from the strictures of having to work themselves to the bone to eat, will produce plays, will produce symphonies.

I don‘t get the point to link will produce symphonies to the wikipedia page of Philip Glass. It is well known that he had to work as a plumber for decades.

qrios··on Why I'm leaving OpenAI and what I'm doing next
This one?

https://www.gurwinder.blog/p/the-outrageous-rise-of-neotoddl...

qrios··on The deconstructed Standard Model equation
The original TeX representation of the formula was written by Thomas D. Gutierrez in 1999 [1]. It was discussed many times on HN, initially in 2016 a day after the post of this article on symmetrymagazine [2].

[1] https://www.tdgutierrez.com/

[2] https://news.ycombinator.com/item?id=12182230

qrios··on Always have a comma in your password, if part of a leak it will break the CSV
What about \""\t\n"?
qrios··on The most cited authors in the Stanford Encyclopedia of Philosophy
> Each author is only counted once per headline entry (subentries are excluded). In 2010, I found that this generated more plausible results than counting authors multiple times per entry.

In my subjective experience, a philosophical text will deal with a specific topic and name it in the title and less often the author. Papers with an author's name in the title often deal with the entire work or a phase of creation. Author names also often appear together with other names "Kant and Hume on Morality"[1].

So the list is for sure not about "… Most-Cited … Authors …".

[1] https://plato.stanford.edu/entries/kant-hume-morality/

qrios··on OpenAI co-founder John Schulman says he will leave and join rival Anthropic
Let‘s see if this recipe will make it into Claude or ChatGPT in two to three years. set a reminder
qrios··on [dead]
https://archive.is/V3Tun
qrios··on Chinese tech companies push staff to the limit
https://archive.is/9ZO1k
qrios··on Detecting hallucinations in large language models using semantic entropy
For the first step CYC[1] could be a valid solution. From my experience I whould call it a meaningful relation schema for DAGs. There is also an open source version available [2]. But it is no longer maintained by the company itself.

[1] https://cyc.com

[2] https://github.com/asanchez75/opencyc

qrios··on SEQUOIA: Exact Llama2-70B on an RTX4090 with half-second per-token latency
The linked github repo [1] seems to have the code available and well documented.

[1] https://github.com/Infini-AI-Lab/Sequoia/tree/main/Engine

qrios··on Just got doxxed to within 15 miles by a vision model, from only a single photo
Same here: Picture from rural Austria in the North and it comes up with Hungaria. Location is in the center of Budapest, 300km distance to the original view point. But architectual style looks really hungarian.
qrios··on Astronomers discover Milky Way’s biggest stellar black hole
„Size“ is misleading. 33 solar masses is in fact small for a black hole. Sigatarius A is still the biggest in the Milky Way.
qrios··on How do machines ‘grok’ data?
A good read if the How and Why NNs are working is of interest. Also the HN discussion [1] about this paper [2] is related from my point of view.

[1] https://news.ycombinator.com/item?id=40019217

[2] "From Words to Numbers: Your Large Language Model Is Secretly A Capable Regressor When Given In-Context Examples" https://arxiv.org/abs/2404.07544v1

(Edit: format)

qrios··on [dead]
Original article in German: [1]

[1] https://www.heise.de/news/Google-Fonts-Abmahnung-Hotel-erwir...

qrios··on DBRX: A new open LLM
Instead of spending x by 10^7 of dollars, Databricks could buy databricks.ai, it's for sale.

But really, I prefer to have as many players as possible in the field of _open_ models available.

qrios··on Conscious AI Is the Second-Scariest Kind
https://archive.is/W50mG
qrios··on Devin: AI Software Engineer
I totally agree!

And I'm sure the reason for that is the garbage input. From time to time I have to perform quantitative code analyses in our so called enterprise repositories. And the results are shocking every time. I have found an extremely poor SQL code block to type cast many columns in hundreds of projects. It was simply copied again and again even though the casting was no longer necessary.

The training base should be sufficiently qualified (and StackOverflow ranking is obviously not enough).

But unfortunately it's probably too late for that now. Now inexperienced programmers are undercutting themselves with poor AI output as training input for the next generation of models.

qrios··on Claude 3 Opus has an issue – Repeats same response over and over
Not using Claude but this behaviour comes not from the model itself. Basically the model does not now what it provided in the other iterations of your for loop. Ask for 20 different profiles. Or, even better: ask for different types of cancers and iterate this list.
qrios··on Building the New Hardware for AI Model Inference [video]
Interesting approach to load the model direct. But no explanation on the video or on the company page[1] about the processing units itself. The numbers from the page (2.3x faster, 1/2 the power) are plausible for top notch FPGAs.

[1] https://www.positron.ai/product

qrios··on Pg_vectorize: Vector search and RAG on Postgres
From our experience simple RAG is often not that helpful as the questions itself are not represented in the vector space (except you use an FAQ dataset as input). Either a preprocessing by an LLM or specific context handling needs to be done.
← PreviousPage 4 of 5Next →