1. Install Ollama
2. ollama pull hf.co/mradermacher/VibeThinker-1.5B-GGUF
3. ollama run hf.co/mradermacher/VibeThinker-1.5B-GGUF

Short answer: em-dashes are used in a lot of professional writing. It's picked up in training, and spread: https://en.wikipedia.org/wiki/Dash#Usage_2

Long answers... I don't really know?

But first some background. LLMs are trained in many stages, with the first being raw text. These 'first stage' LLMs don't behave like chatbots you know; they take a raw passage of text and simply try to complete it word by word, with nothing both the text's context to go on. They're (generally) unbiased, as they've simply been trained on billions of pieces of work with no preference for any particular one.

The subsequent training (to get them to behave more like question/response chatbots) is where they pick up 'style' and quirks, and where trainers may use smaller, more deliberate datasets, and this is where they (specifically OpenAI models, which have this habit in particular) likely pick up em-dashes from. Maybe it's neatly formatted textbooks or summaries, maybe it's reformatting (kind of like how your phone will turn neutral ' " ' quotes into left/right ones), I don't know... But it's there in that stage of training.

They also do what's called "RLHF" at this stage, where they 'pick' answers real humans prefer, and it might be possible that the human testers tend to prefer the neatly formatted em-dash answers more frequently. And all this creates a feedback loop where the more an LLM outputs an em-dash, the more it 'learns' to do it.

...There's also another factor.

It's an open secret that LLMs train on each other's output, deliberately.

As a consequence, they pick up each other's habits. It's possible to even map their "slop patterns" into taxonomical trees, which EQBench attempts to do though its tests:

https://eqbench.com/creative_writing.html

Any notice funny patterns, like how Kimi K2 switched to 'copying' Claude instead of Gemini for its thinking model.

Hence the em-dash habit has spread to other LLMs kinda like slang spreads between human groups.

I wonder how I can try this (a bit new to this llm stuff).

I think a factor is forced developer adoption with mandates coming from senior management and business leaders who have been sold on the more code for less investment pitch without understanding this level of technical nuance.

His experiences reveal a pattern: AI can rapidly produce 70% of a solution, but that final 30% – edge cases, security, production integration – remains as challenging as ever. Meanwhile, trust in AI-generated code is declining even as adoption increases.

I'm very much intrigued by this contradiction where where adoption of AI is increasing, but the trust in the code it generates is declining. Is it a case of the more developers use AI coding tools, the more they become aware of the shortcomings and problems?

Hey Trump what do you think about climate change?

Well the weather is nice in Florida this year, you know Florida? Believe me, the best state in the country. Let me tell ya folks, nobody has weather like we do in Mar-A-Lago. Huge weather down there, unbelievable. *tiny accordion hands* But Israel doesn't have that much weather because of Obama and Crooked Hillary...

On Friday, reporters asked Trump if he was concerned about an "AI bubble." 

"No, I love AI," Trump said. "We're leading China, we're leading the world."

Ha ha!

That doesn't seem possible

This is what should've been in the description, imo

The new methods developed by DeepSeek (and published in its latest paper) could help to overcome this issue. Instead of storing words as tokens, its system packs written information into image form, almost as if it’s taking a picture of pages from a book. This allows the model to retain nearly the same information while using far fewer tokens, the researchers found.

Besides using visual tokens instead of just text tokens, the model is built on a type of tiered compression that is not unlike how human memories fade: Older or less critical content is stored in a slightly more blurry form in order to save space. Despite that, the paper’s authors argue, this compressed content can still remain accessible in the background while maintaining a high level of system efficiency.

Humans aren't perfect either, and often spew confidently wrong information.

Ummm... Duh? But I guess at least we have a another paper to cite when arguing against delusional claims about LLMs.

Possibly because it's very NVIDIA focused?

I'm excited about the prospect of small models e.g. fine-tuning against MCP calls to give APIs language interfaces.

There's a lot of frustrated zealots on lemmy. They prefer faulty ideology over pragmatism. Slowly getting angrier as their world view and reality diverge.

I'm confused why people are voting down an article about AI in an AI community, discussing small language models, which are much better in terms of energy consumption and the environment.

Stupid headline is about as useless as could be.

ChatGPT Atlas is a Mac-only web browser with a variety of ChatGPT-enabled features.

(approximately $1.37 billion)

Vulkan?

I'm not too familiar with either, but this article goes into more detail: A Comparison of Modern Graphics APIs

Interesting. Is that similar to Vulkan?

Ollama uses the Metal API on Apple Silicon Macs for GPU acceleration.

Wat I don't understand is how it's supported on Mac. Do they have NVIDIA cards? I thought everything was dominated by CUDA. Did the lock-in company implement CUDA?

The prospect is definitely tempting, but the problem is that the market can remain irrational for longer than I can remain solvent...

What I got from the article is an example of how generative AI can fix a bug, if you provide it with a reproducing case. Yet funnily enough, the AI introduced a bug in the first place by using an older version of a dependency.

Wth all these predictions, I wonder how many people shorted their stock. Once this crashes (if it crashes), some people will get very rich!

Perhaps I'm understanding it wrong, but isn't agentic search pretty much the "retrieval" part of retrieval-augmented generation (RAG)?