pilotneko

Businesses raise prices to cover the interchange fee, which impacts lower income consumers disproportionately.

Where did you see a F-150 Lightning new for less than $40k anywhere?

Dollar General and Family Dollar are smaller stores that are generally the only option within a reasonable travel distance. Here in the South, you might be able to catch a bus to Wal-Mart, but it’ll take 2-3X more time (1 hour instead of 20 minutes), so people go with the closer option even thought it is more expensive. No guarantees that Wal-Mart will be cheaper either.

I mean, it’s not trivial. There is a lot of work involved with enabling tool use at scale so that it works most of the time. Hiding that work makes it worse for the common user, because they aren’t necessarily going to understand the difference between platforms.

That’s because LLM’s generally don’t cite their sources. Web search is a tool outside of the LLM. Depending on the particular chat interface, there are any manner of tools in place to augment LLM capabilities/outputs, and they are constantly changing.

If you read the model card, Qwen3-Next can be extended to 1M context length with YaRN.

> Qwen3-Next natively supports context lengths of up to 262,144 tokens. For conversations where the total length (including both input and output) significantly exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively. We have validated the model's performance on context lengths of up to 1 million tokens using the YaRN method.

Source: https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct#proc...

Sometimes you are searching for supporting evidence that is semantically related. COPD was just an example, you won’t get a direct keyword match if the Physician is searching for lung disease.

Keyword alone sucks for negation. Searching a set of patient documents for “Which of my patients has COPD?” to get a set of responses that states “COPD not indicated” will not endear you to the physician using your tool. Hybrid (keyword + semantic) is much more all-encompassing.

I use POE for my outdoor security cameras. They do not have access to a dedicated plug on the outside of my house.

I mean, just this week there was a popular post “My AI Skeptic Friends are all Nuts” that generated a ton of such discussion and dismissal. https://news.ycombinator.com/item?id=44163063

It is common to use long context embedding models as a feature extractor for classification models.

Likely a very small one, or none at all. It’s 450M parameters, which is still in the sweet spot for CPU inference.

I think that people are uncomfortable with the idea that Google Maps is centralized and can unilaterally change what you see. Having an offline version of a map helps protect against sudden change (go forward or retroactive).

As an aside, I do really like organic maps. I keep it installed with downloaded maps for when I travel to places with poor signal, including hiking trails.

I have used the Remarkable 2 for papers, but it is slightly too small to read text comfortably. I’m also an active reader, so I miss the color highlighting. Annotations are excellent. For now, I’m sticking to reviewing papers in the Zotero application on my iPad.

Well, that seals the deal. If it doesn’t work for dentists in Japan, there is no point changing anything at all for US healthcare policy.

I don't use Mistral 7B alone, this is just a component in a RAG-based system. A system that is 1) not clinical facing, 2) not used in clinical decision making, 3) provides in-line references sources for end users to validate information, and 4) is inherently human-in-the-loop.

Mistral 7B has a context window of 32k. I use it in production for medical summarization tasks supporting appeals and physician advisory services.

I acknowledge there are issues on the provider side, but it is disingenuous to say that providers set the prices alone. Payers introduce a ton of inefficiencies in billing and also remove money from the system, which negatively impacts care. They implicitly affect care patterns and pricing through denials.

HMOs, for all their problems, have many advantages as well, such as the aligned incentives you allude to.

Don’t forget “we don’t build weapons…well the company does but my team doesn’t.”

Every release is novel. Once something has been around for a while and is still being referenced, you know it’s worth learning.

Waiting 3-6 months to take a deep dive is a good pattern to prevent investing your time in dead-end routes.

I mean, the ones that do ask don’t proceed to signing up. I think we are seeing a form of survival bias.

It was a bit surprising to click this link on my VPN. Wish it had a NSFW tag. Uncensored is a bit of a crapshoot when it comes to GenAI, half of the time is just means "no instruction tuning."

No worries. Sorry if I came off as combative.

I am aware of the replaceable nibs, and I don’t have a problem with those. I have a Remarkable 2 and I use it practically every day. I replaced the nib with a titanium one and it feels very similar to a ballpoint pen writing on a clipboard.

The new marker requires charging. It is not compatible with the Remarkable 2 or other e-ink tablets like the Kindle Scribe. Look at the features on the store page, both the Marker and Marker+ talk about wireless charging.

Also, from their FAQ about the Marker:

Will other digital pens work on reMarkable Paper Pro? No. The new Markers have been custom-built for reMarkable Paper Pro’s Canvas Color display. Other digital pens, such as the Markers for reMarkable 2 or those developed by third parties, aren’t compatible with reMarkable Paper Pro. For more details, see our compatibility guide.

I have been really interested in this device, but the proprietary stylus really turned me off. In addition to not working on any other e-ink reader/notepad, it is a mystery to me why they went with a stylus technology that requires charging.

Sorry, life got busy and I haven’t been able to get back to you. I was referring to pipelines in the Transformers package from Hugging Face. https://huggingface.co/docs/transformers/v4.45.2/en/main_cla...

These are essentially function calls for you to run pre-trained models. If you want to continue this conversation elsewhere, feel free to shoot me an e-mail. It’s just my username @ gmail.

Hugging Face does maintain a package named Text Embedding Inference (TEI) with GPU/CPU-optimized container images. While I have only used this for hosting embedding models, it does appear to support Roberta architecture classifiers (specifically sentiment analysis).

https://github.com/huggingface/text-embeddings-inference

You can always run a zero shot pipeline in HF with a simple Flask/FastAPI application.

It shields you from getting your debit card skimmed. The credit card is also easier to contest if there is an issue. If you pay your credit card balance off every month, you never end up paying interest either. Finally, you can accrue points and rewards spending money you already planned to spend.

That's weird, because the repository states Apache 2.0. https://github.com/THUDM/GLM-4/blob/main/LICENSE

Oops, you are right. The code is Apache 2.0, license for the model weights is separate.

I experimented with this model and vLLM around a month ago. The long context length is attractive, but it was incredibly slow on a g5.12xlarge (4 NVIDIA A10G GPUs). I actually could not get it to respond for single examples longer than 50K tokens.