The site looks vibe coded. It’s possible they just bought the trademark, quickly set up a site and hope to capitalize on the reputation of former Twitter.
The $20 and $40 early access certainly makes it seem like it. Even if that isn’t the case, time has shown that centralized social media has flaws so idealists will just stay with Mastodon while the people that want a large user base and easy of use will just use BlueSky instead. IMO there is no market aside from a possible gofundme-invention type exit scam.

I feel like everyone knows it’s a bubble but that people are expecting it to be like the Dot-com bubble where a small amount of companies succeed, hoping that they invest in those companies.

Network effect / discoverability and free CI time

True, I understand why this is better than all those companies scraping it individually (for both ability to opt out and site load), but the way they handled opt out is still quite silly.

I checked the repo but there’s no code? Just some HTML etc.
It seems like they just plan for this thing to exist but it doesn’t actually exist yet?

EDIT: Their explanation:

This repository is currently in its early setup phase.

At the moment, the repository mainly contains project scaffolding, policy files, and the initial project direction. Firmware sources, board support, build instructions, and flashing guidance are still being prepared.

That means zgm is early, but it is intentionally early in public. The goal is not to wait until everything is finished before opening the door. The goal is to make the long-term direction visible and let the project grow into a useful open firmware platform in the open.

I’m not sure if that means that they still have to make it, or that they just have to finish it before publishing it?

The banner at the top of their site tells a different story

Cryptocurrency projects are no longer allowed. View change

I don’t like these blanket bans, are they going to ban torrent clients next?

It could also be a way to encourage more regulation to push out competition with compliance cost

From my personal experience, most companies don’t look at the benchmarks and simply buy “AI” from a company based on their perception of that company.

Luckily it probably won’t get made and the company will just rugpull their investors

Better than that

And even if they publish the model it’s almost always just open weights

According to artificialanalysis.ai’s latest benchmarks, it scores better than Opus 4.8 set to max, despite costing half as much per task.

It also beats the top models of some of the largest US tech companies such as xAi, Meta, Google, and Nvidia.

I wonder what the US tech investors will think of this, and what this will mean for the financial AI bubble.

You can use it through openrouter

AFAIK, their open models are distributed as weights, not executables and are therefore not able to start network connections / run code. There is of course tool-calling functionality but that just works by having the model output a special pattern and having something external run predetermined commands based on that.

This reminds me of something I sometimes see in shows on like Netflix and other media. I can’t remember a specific example, but you often have generic anti-capitalist comments from characters (often portrayed as edgy). It often feels a bit, artificial, like a “fellow kids” moment but politically, I guess? Like activism as a prop / character trait, inserted into a multi-million media production. Maybe someone else can better put it to words, if I had more time I would’ve written a shorter comment

I agree. The worst part about GitHub training LLM’s on my FOSS code without permission for me is that they then keep the models to themselves. Like if you’re going to use all my code without permission, at least allow me to run the model locally.

My personal opinion is that all models trained on copyleft code should be open-weights, most FOSS licenses didn’t account for this specific possibility, but this is the only way to follow them in spirit.

Deepseek recently published a paper in which they describe that vision tokens contain more information than text tokens and that this can be used to compress context.

We present DeepSeek-OCR as an initial investigation into the feasibility of compressing long contexts via optical 2D mapping.

Experiments show that when the number of text tokens is within 10 times that of vision tokens (i.e., a compression ratio < 10×), the model can achieve decoding (OCR) precision of 97%. Even at a compression ratio of 20×, the OCR accuracy still remains at about 60%. This shows considerable promise for research areas such as historical long-context compression and memory forgetting mechanisms in LLMs.

It reminds me of LLM caveman speak, it used to have another option to use Chinese instead of English. A language like Chinese is seemingly better at encoding information in fewer tokens and I think this is the same mechanism why OCR tokens work so well.

That said, I also doubt that voice messages are more efficient than text prompts, but it’s best not to waste too much time engaging with these sorts of LinkedIn posts (and LinkedIn in general).

It’s being downvoted with relatively little discourse because it’s an insult with no relevance to the topic, in addition to supporting a comment from someone who is either trolling or has no idea what they’re talking about

I think the latter