HDBaseT

If M-DISCs were caught in a fire, I don't think they are readable either.

I think we'll have the hallucination problems solved in a 1000 years. One would hope at least.

If 1000 year AI cannot write firmware & drivers for fairly standard discs, then AI has failed massively.

"Claude Fable 5 and 5.1 are not included in LifeSciBench Gold v1, GeneBench Pro v13, and MedChemBench because they refuse the majority of questions in these evaluations.12"

Sounds about right. Alignment is important, but also being able to do mundane tasks is important too.

In theory, but in practice how difficult is that?

By default, even without the training endpoint the pricing is pretty competitive, especially against Opus and Fable. [1] The 'muse-spark-1.3-contributor' endpoint is by far the cheapest, significantly cheaper per M than ChatGPT Luna, significantly smarter than Luna too.

This price/intelligence beats even legacy DeepSeek V4 Flash pricing.

[1] https://artificialanalysis.ai/#total-cost-tabs

Not to mention, this is hyper competitive against even Chinese providers given its multi-modal support.

Muse Spark 1.3 supports Text, Image, Video, File, Audio inputs. We've only started to see models from China include image and video inputs recently.

A small number of inputs in a large dataset can poison training data pretty drastically. Anthropic wrote a good article about it a while back [0]. This should mean its possible to pull back that information fairly easily.

It is hard to not feed it "secrets" too. Models will see path names, read compose files, etc. Of course you can configure things to not leak this type of information, but its not default in most harnesses and isn't 100% sufficient anyways.

[0] https://www.anthropic.com/research/small-samples-poison

Zuck said Muse Spark 1.2 should be getting open weights "soon" on a tweet from a few weeks ago.

The problem inference providers will not be able to get anywhere near the contributor pricing.

Despite this, Qantas is consistently rated one of the top 3 safest airlines in the world every year.

"I believe we're reaching the upper limits about what generative AI can do and how accurate its outputs can be."

This was said in February 2024. This is months before GPT-4o released. GPT 4.5 was released a year later.

I don't know anyone holding on to models of GPT 4.5 caliber, yet know 4o. ChatGPT 4o and 4.5 score 8 and 14 points on Artificial Analysis benchmarks. [1]

For perspective, Qwen 3.6 27B, which can be ran on single GPU setups, scores 3-5x that on modern benchmarks.

I don't know why anyone would even make a claim like that in the first place. It's like saying "Computers are never going to get faster". I feel like we could run out of sand and still have faster machines over time. Just a silly thing to say.

[1] https://artificialanalysis.ai/?models=muse-spark-1-2%2Cgemin...

It does raise the question, what percentage of programmers are predominantly writing code by hand? Is it higher than 50%?

The pricing on Luna is just insanely good. Haiku simply cannot compete at almost any intelligence level against DeepSeek V4 Flash or Luna.

The private tracker elitism really needs to stop. The sheer amount of mod power trips and fake elitism is really quite annoying.

This is almost never worth it with most sites and generous bonus points.

If you see a few TBs on a Seedbox or your server for a year, you are pretty setup for life on most sites.

Apple was also allegedly "caught off guard" by the Macbook Neo demand.

I don't really see how they couldn't see the Local AI demand or demand for a cheaper Macbooks. This just reads like marketing imo.

false equivalence.

When I break into peoples houses, I tend to take the NVR with me.

I just don't think Matrix is the solution.

[1] https://benharri.org/matrix-sucks/ [2] https://telegra.ph/why-not-matrix-08-07 [3] https://piegames.de/dumps/a-week-in-matrix/

The systemic change is to prevent lobbying from the likes of Israel.

This event has nothing to do with climate change and is pure fear mongering.

This is the same adage used by the Governments with the "think about the children" campaigns, its a distraction. A cop out.

Sounds cool, please document your steps and pitfalls. I would live to try this one day.

DNS is important and 100TB of RAM is effectively nothing.

This is true, although Apple could of just as easily prevented alternative OSes as a whole if they wanted to.

I2P Nitter instances exist: va7j3ssqh7e26twc2sflrwjrn7uxryuenatcmfp7njnr5hmvtuiq.b32.i2p

Ox Alpha was also serving 10T+ tokens a day for free.

When it first launched on OpenRouter I was getting nearly 70 Tokens/second.

I only use Luna (max), I find it very rarely just reasons. In fact, I find it reasons too little.

I assume a Aquantia one.

Who is going to order DoorDash?

The Work from Home programmers surely aren't, since they've been displaced.

I spoke two a friend last night about this, he had no idea what "tokenmaxxing" was and had never heard of "vibecoding".

We truly do have distorted view on anything outside our bubble.

A highly modular plugin system would solve this though. A great base product, with plugins (both community & official) which could expand feature sets.