Hacker News Reader

ru552on DeepSeek proves the future of LLMs is open-source

It's been confirmed to run on a machine with no internet access. So it isn't reliant on external requests, though it could still be trying to make them.

ru552on DeepSeek proves the future of LLMs is open-source

With an LLM, the actual 0s and 1s of the model are fairly standard, common, and freely available to anyone that wants to use them. The "source code" for an LLM, is the process used to create the outcome, and to an extent, the data used to train with. DeepSeek released a highly detailed paper that describes the process used to create the outcome. People/Companies are actively trying to reproduce the work of DeepSeek to confirm the findings.

It's more akin to scientific research where everyone is using the same molecules, but depending on the process you put the molecules through, you get a different outcome.

ru552on Instagram deals reveal Meta is offering TikTok creators as much as $300k to post

Of course it's legal. They offer you money to exclusively post on their platform. Joe Rogan got paid to exclusively podcast on spotify. All of the talking heads on TV news are paid to exclusively be on their specific channel (ABC,NBC,FOX,etc.) The twitch streamer, Ninja, that got paid to move from twitch to Microsft's failed Mixer platform, etc. etc..

ru552on Orion, Our First True Augmented Reality Glasses

"I think we still have some room to grow there in terms of aesthetic for the majority though."

So does Meta and it's the reason they aren't actually releasing this as a product. They just showed us where they're at and gave us an idea on where they want to go in the next few years.

ru552on Update on Llama adoption

>They all use the same model and the same transformer algorithm. The model has an EULA, you need to apply for downloading it, the training data set and the training software are closed.

Everything in that sentence is false except the training data part.

>So the "competitive" system means that everyone uses LLama and PyTorch.

This sentence shows you don't understand the LLM landscape and it's also false.

>Sounds really open

Correct. They partner with practically every vendor available for inference, which, isn't even needed if you run their models locally.

Meta has done a lot of wrong things over the years. How they are approaching LLMs is not one of them.

ru552on The journey of an internet packet: Exploring networks with traceroute

~~ Security through obscurity does work.

As a layer.

ru552on The journey of an internet packet: Exploring networks with traceroute

Wait until DNS comes up again. I don't know how many DevOps, SRE, Senior Architects, etc. that I've worked with that just don't understand how names are resolved. There's a wide array of specialists that don't understand general fundamentals.

ru552on NASA acknowledges it cannot quantify risk of Starliner propulsion issues

He's probably talkin about how Boeing is multiple years late and still has a shoddy product.

ru552on CriticGPT: Finding GPT-4's mistakes with GPT-4

Anthropic has attributed Sonnet 3.5's model improvement to better training data.

"Which data specifically? Gerstenhaber wouldn’t disclose, but he implied that Claude 3.5 Sonnet draws much of its strength from these training sets."[0]

[0]https://techcrunch.com/2024/06/20/anthropic-claims-its-lates...

ru552on US EPA Enforcement and Compliance on Apple Fabrication

She's not a reporter. She was fired from Apple for breaking an NDA. This isn't the first time she's made sensational claims against Apple.

ru552on Apple Intelligence for iPhone, iPad, and Mac

This is exactly what they are going for. You can just ask Siri now "what day did my wife say the recital is?" and Siri spits the answer out without requiring you to go scroll through your messages. Who cares that an LLM did the work?

ru552on Apple Intelligence for iPhone, iPad, and Mac

you have to acknowledge a pop up authorizing your request be sent to OpenAI every single time it happens. it's not going to happen by mistake.

ru552on Apple Intelligence for iPhone, iPad, and Mac

~It seems like it will require M1+ on Macs/iPads, or an iPhone 15 Pro.

They specifically stated it required iPhone 15 Pro or higher and anything with a m1 or higher.

ru552on Systematically Improving Your RAG

*What makes it so useful?

One example is in finance, you have a lot of 45 page PDFs laying around and you're pretty sure one of them has the Reg, or info you need. You aren't sure which so you open them one by one and do a search for a word, then jump through a bunch of those results and decide it's not this PDF. You do that till you find the "one". There are a non trivial amount of Executive level jobs that pretty much do this for half of their work week.

RAG purports to let you search one time.

ru552on Carmakers Will Give Your Location to Police Without a Warrant, Senators Say

It's not illegal, it's just that it doesn't allow it in the UI. Who do you call to have that service disabled? No one knows. The best you can do is talk to someone and have the auto renewal turned off. That doesn't turn off the existing service until the renewal period hits though.

ru552on Ilya Sutskever to leave OpenAI

More like they iced him for the last 6 months to ensure he wasn’t taking their lead to a competitor. He probably hasn’t touched anything in that time.

ru552on Chime Design and Build

Came here thinking it was the Saas Chime. Delighted it was the physical chime.

ru552on Snowflake Arctic Instruct (128x3B MoE), largest open source model

To stay fair, their "17B" model sits at 964GB on your disk and the 70B Llama 3 model sits at 141GB. unquantized GB numbers for both

ru552on Snowflake Arctic Instruct (128x3B MoE), largest open source model

the unquantized model is almost 1tb in size and the benchmarks provided by Snowflake shows performance in the middle of the pack compared to other recent releases.

ru552on Snowflake Arctic Instruct (128x3B MoE), largest open source model

Abnormally large. I don't see the cost/performance numbers going well for this one.

ru552on Meta Horizon OS

With this and Llama, looks like Zuck is trying to be the Linux of whatever the next big wave is

ru552on Apple has reportedly acquired Datakalab

Not if the acquired isn't public.

ru552on Mixtral 8x22B

a macbook with 64g of ram

ru552on Mixtral 8x22B

easiest is probably with ollama [0]. I think the ollama API is OpenAI compatible.

[0]https://ollama.com/

ru552on Mixtral 8x22B

Today. Might change tomorrow at the pace this sector is at.

ru552on Mixtral 8x22B

There's a large amount of liability in disclosing your training data.

ru552on How Chain-of-Thought Reasoning Helps Neural Networks Compute

"It's a very misleading statement to describe today's LLMs as "simply predicting the next word" even if it's in part true."

That's exactly what they do. Don't anthropomorphize a computer program.

ru552on The idea that AI will just augment jobs never replace them

I'd argue that jobs wont be lost so much as change, just like all the previous technological jumps. You had an entire industry to cater to horse travel until cars came along. Then you needed an entire industry to cater to cars and their maintenance.

ru552on Machine Learning Is Still Too Hard for Software Engineers

https://fleuret.org/francois/lbdl.html

There's a 0-60 in 3 seconds one. Specifically about deep learning, which is a subset of ML. Deep learning is what is used to build LLMs ala ChatGPT.

ru552on Harvard President Claudine Gay resigns

Harvard is private, but they do receive public funds. This exposes them to additional scrutiny.