wskish

Skills are basically (on-demand) man pages for other libraries or cli tools, or SOPs for other more complex processes. A beautiful, minimal abstraction.

codebases of high quality open source projects and their major dependencies are probably another good source. also: "transformative fair use", not "stolen"

I think about this a lot. Consider the difference between the tidy signage of Tokyo versus the pell-mell streetfronts of Hong Hong. Societies should be able to choose how businesses impinge the public space.

Amplified messaging from corporations is not the same as the free speech of individuals. Just as we disallow advertising for cigarettes and hard liquor on TV, a democratic society should be free to select other classes of messages that corporations are not permitted to amplify into public spaces.

"Let’s delve into the details"

To fully experience the night sky there is no substitute for getting up to altitude. Most of the atmosphere that really interferes with the view is way down low. Try a moonless night at 8000 to 11000 feet elevation in the eastern Sierra or Yosemite high country to really blow your mind. Boss level is a summit bivy during a new moon meteor shower with a close friend or lover.

"The Andrew Message System is one of the most ambitious and technically successful systems yet built in the area of electronic communication. In this paper, the authors of the system explain the key decisions in the architecture and reflect on what was done right and what was clearly wrong. Implementation details are mentioned only in passing, in order to maximize the relevance of this paper for the designers of successor systems."

I came here to mention MenuMeters as well. It's the closest thing I can find to those beloved X window system load indicators of yore.

interesting, GPT refuses to summarize this content: "I'm sorry, but I can't generate a summary for that content." per https://github.com/jiggy-ai/hn_summary & https://t.me/hn_summary

I'm curious about the methodology but couldn't find an associated paper or research report. The original source seem to be university PR.

It is amazing much room there is to re-optimize cities around smaller electric vehicle transportation when you think on a slightly longer timescale.

It isn't so clear to me. If it conveys a human in any way that humans don't naturally move unassisted then in my mind it qualifies as a conveyance and hence a vehicle, especially in the sense of the the French véhicule, from Latin vehiculum (“a carriage, conveyance”), from vehere (“to carry”)

"Attention with Linear Biases (ALiBi): Inherited from the MPT family, this feature eliminates the context length limits by replacing positional embeddings, allowing for efficient and effective processing of lengthy documents. In future we are planning to finish training on our larger dataset and to increase amount of tokens for context."

This is interesting, but also confusing. What is the current context limit? It mentions eliminating the limit but then mentions increasing it in the future.

here is code (with several examples) that takes it a couple steps further by validating the output json and pydantic model and providing feedback to the llm model when it gets either of those wrong:

https://github.com/jiggy-ai/pydantic-chatcompletion/blob/mas...

Do they mention anywhere the definition of "low quality" data or the proportion of removed data that was low quality versus duplicate?

They mention "When upsampled, we expect SlimPajama to perform equal to or better than RedPajama-1T when training at trillion token scale." But i guess "upsampling" in this case is just explicit duplication of the training data. So the only potential gains would be from the removal of the low quality data?

Interesting that we have OpenAI using the more generic "GPT" here. Previously they would refer more specifically to ChatGPT, GPT-3, or GPT-4. I am guessing this is related to their trademark application for GPT, which was initially refused by the USPTO on the grounds of "GPT" being "merely descriptive".

https://news.ycombinator.com/item?id=36155583

I uploaded the USPTO office action to a JiggyBase (https://jiggy.ai) collection and used GPT-4 and the JiggyBase ChatGPT plugin to help make sense of this:

https://chat.openai.com/share/6a16cc77-4ef7-4e3a-bddb-dedd3f...

"REFUSAL—MERELY DESCRIPTIVE: Registration is refused because the applied-for mark merely describes a feature, function, or characteristic of applicant’s goods and services."

There are other details that exceed my trademark ken. Anyone with trademark experience care to parse this for us?

In this case we mean you can get some of the benefits of langchain and pinecone such as semantic search and augmented GPT retrieval without needing to deal with vectors, chunking, and llm tooling at such a low level. You can upload docs and then begin chatting against them immediately. JiggyBase is just a higher level abstraction on top of these same type of components which may be useful in a lot of cases where you don't need full control over the vector embeddings and such and just want to interact with your data.

For anyone thinking about applications of langchain and pinecone but who are looking for something more turn-key check out https://jiggy.ai

The core is actually open source as well, allowing you to take your data back out via sqlite and hnswlib (https://github.com/jiggy-ai/hnsqlite)

My partner and I recently launched something similar we are calling JiggyBase:

https://jiggy.ai

The core is open source (https://github.com/jiggy-ai)

this page presents as bottom 2 to 10% of internet

based on my experience with GPT-4 in coding tasks it responds extremely well to just appended raw error output text without additional explanation.

yes thats actually it

After working through several projects that utilized local hnswlib and different databases for text and vector persistence, I integrated open source hnswlib with sqlite to create an embedded vector search engine that can easily scale up to millions of embeddings. For self-hosted situations of under 10M embeddings and less than insane throughput I think this combo is hard to beat.

https://github.com/jiggy-ai/hnsqlite

Here's 40 lines of python code that I've found to be unreasonably effective at accomplishing something similar:

https://github.com/jiggy-ai/pydantic-chatcompletion/blob/mas...

I wish more hardware products supported the "embedded airtag" feature enabled by the Apple Find My program that enables 3rd party manufactures to effectively embed airtags into their products.

https://mfi.apple.com

It was the "petition to make special" (something like a request to speed up the process) that was dismissed, not the trademark application.

https://tsdr.uspto.gov/documentviewer?caseId=sn97733259&docI...

It is interesting how at this point in time the OpenAI -> Nvidia -> TSMC -> ASML tech stack is effectively single source all the way down. We haven't really seen anything like that in the prior semiconductor, internet, and mobile tech waves.

Those standing on the shoulders of giants are effectively very tall.