modernpink

I, and many economists, would argue that the price signal, profit and "serving society" are more tightly interwound than many like Hideyuki are mislead to believe.

Please tell me there's a way to stream YouTube videos in 256 colors in the terminal as ASCII art. As God intended.

China is outcompeting Germany as a manufacturer of high value goods, particularly cars [1][2]. Last year it overtook Germany as the second biggest car exporter. In the first three months of this year, China has surpassed Japan to become the world's top car exporter for the first time. This is due to its strength in EVs (while Germany's strength is in ICEs). Volkswagen holds a 2% share of the Chinese EV market while the Chinese BYD dominates with 40% (Tesla in second at 10%) [3]. It won't be long until Europe imports more cars from China than China imports cars from Europe. If you had said this only a few years ago most Europeans would have laughed in disbelief.

With a world moving away from ICE to EV, and with Chinese brands winning at home and abroad, it presents a double whammy nightmare for Germany. Lowered exports to China, plus lowered exports in other markets. And this is all on top of energy prices from the fallout of the conflict in Ukraine, the shut down of nuclear power thanks to the Greens and the geopolitical tensions between the West and China.

While Germany's economic model has served it well during an era of a rising China, it may not be the case in a era of a risen China.

[1] China’s carmakers outstrip foreign brands in its electric vehicle boom https://www.ft.com/content/dd149923-1c5d-4e5e-8be6-d979aa48a...

[2] Big drop in German exports to China raises fears over EU’s economic powerhouse https://www.ft.com/content/3034d39c-118a-44d3-b9bf-6482b7860...

[3] Can Volkswagen win back China? https://www.ft.com/content/afad4566-6404-483e-934f-c905507bf...

The reality is that the quality of production in those countries isn't going to match China, that is why things are still Made in China despite rising wages and geopolitical tensions.

The FT reported that 50% of Indian iPhone _casings_ did not pass quality checks [1] to be sent forward for factory use. That is just the casing, a not particularly complex piece relative to other iPhone parts. On top of that, Mac Pro production moved from the US back to China [2] after an attempt at "Made in America" due to easier parts sourcing.

I can see Vietnam getting there, but again the scale is just not comparable to China.

[1] https://www.ft.com/content/0d70a823-0fba-49ae-a453-2518afcb0...

[2] https://www.ft.com/content/0cddd49e-99d2-11e9-9573-ee5cbb98e...

What are the commercial applications of mega context window LLMs at current prices? I would guess mainly legal. And what strategies would you rely on to reduce the accumulating costs over the course of a session?

What was the dollar cost to do this work? To iterate over a 40k context must be expensive.

* is all you need

Pareto principle. With the dawn of AI systems and the scale they require I wonder how much more unequal things will get. Previously you would have no issues running the state of the art CIFAR classifier model on your laptop, now running your own semi-state of the art LLM is not even worth doing for many companies.

Yes, that was my point.

I tried two of the "failure mode" examples given: The Russian bears in space, and walking across a river with an average depth of 3 feet. The examples fail in GPT-3.5 while GPT-4 gives the "correct" response. The article is already out of date.

Well for one thing the US seems committed to blow up TSMC to prevent China getting the tech.

https://www.theregister.com/2023/03/14/us_china_tsmc_taiwan/

Market competition and innovation in both ML and hardware has consistently driven down the price of AI in the past decade. You only have to look at where we are with capabilities today compared to ten years ago when CIFAR100 classifiers were the state of the art.

Barring a Chinese invasion of Taiwan, these APIs will halve in price over the next year.

Or rather, more analogously, a self-driving car that has a range of 10 000 miles but sometimes makes mistakes when driving vs a self-driving car with a range of 800 miles that never makes mistakes. Once you've have a taste of intelligence it's hard to give up.

However, in many applications there is a limit on how intelligent you need the LLM to be. I have found I am able to fall back to the cheaper and faster GPT-3.5 to do the grunt work of forming text blobs into structured json within a chain involving GPT-4 for higher-level functions.

I believe this goes back to the ancient Shinto belief that all beings are inhabited by kami (gods). Therefore to pierce a food is to pierce the god within. This is also why origami developed more broadly in Japan, because paper cutting (a Chinese artform) would damage the kami whereas folding fortified the kami. That is why much origami is of animals. There is also a belief that all food prepared in Japan is given "as is": a crafted piece given down from a master sushi chef, a wagashi artisan, or a tonkatsu expert. Sliced, shaped and formed as a bespoke morsel for the diner, not to be tampered with, but to be appreciated whole.

Is this anything like what Langchain supports which is various strategies for chat history compression (summarisation)?

https://www.pinecone.io/learn/langchain-conversational-memor...

But aside from this, if I have a large document that I want to "chat" with, it looks like I am either chunking it and then selectively retrieving relevant subsections at question time, or I am naively dumping the whole document (that now fits in 32k) and then doing the chat, at a high cost. So 32k (and increased context size in general) does not look to be a huge gamechanger in patterns of use until cost comes down by an order of magnitude or two.

Interesting. Have you tried StableVicuna?

GPT3 is dated so many open source models are competitive with it, but Vicuna 13b is supposed to be competitive with GPT4

At $0.60 for 20k prompt tokens, it's not going to be cheap so they will need to bring the price down to get broader adoption.

As far as I can tell, the initial reading in of the document (let's say a 20k token one), will be a repeated cost for each subsequent query over the document. If I have a 20k token document, and ask 10 follow-up prompts consisting of 100 tokens, that would take me to a total spend of 20k * 10 + (10 * 11)/2 * 100 = 205,500 prompt tokens, or over $6. This does not include completion tokens or the response history which would edge us closer to $8 for the chat session.

Yes that's imprecision on my part. For a table D is fixed, but not necessarily across tables in the vector database index

An embedding model will map a string of text (of variable length) to R^D. Each model has its own fixed D, yes. (Typically the vector is a unit vector for performance reasons.) The main function of vector database is similarity lookup so you would calculate approximate nearest neighbours of a vector with a scalar vector distance metric (e.g cosine similarity). These similarity metrics operate on two vectors of the same dimension.

You would not mix and match embedding models (e.g. with differing dimensions) at look-up time. The target vector table assumes you will look it up with a vector created from the exact same embedding model and version that was used to backfill it.

The API documentation for a look up operation may be more illuminating here:

>vector (array of floats)

>The query vector. This should be the same length as the dimension of the index being queried. Each query() request can contain only one of the parameters id or vector.

https://docs.pinecone.io/reference/query

Not sure what this comment is driving at. "At it's core" a database is a collection of data, sure.

You can sell the A100 after once you're done as well. Possibly even at profit?

No it doesn't. I find its evaluation rather fair.

Ironically the GPT4 translation is better

>Both translations effectively convey the main points and ideas of the original Italian text. However, the translation I provided seems to be more fluent and coherent, using more natural English phrasing and terminology. For example, "the Guarantor for the protection of personal data" is translated as "the Italian Data Protection Authority," which is more common and clearer in English.

>While Google's translation is generally accurate, it has a few awkward phrases or word choices that make it slightly less clear or idiomatic, such as "the best known of the relational artificial intelligence software" instead of "the most well-known relational artificial intelligence software."

>Overall, the translation I provided is more polished and reads more smoothly in English, which may be preferred for better understanding and clarity.

>Taking into account accuracy, fluency, and clarity, I would rate the translations as follows:

>My translation (ChatGPT): 95/100

>- The translation is accurate, fluent, and clear. It effectively conveys the original text's meaning and reads smoothly in English. The phrasing and terminology used are natural and idiomatic.

>Google's translation: 85/100

>- The translation is mostly accurate, but there are some instances of awkward phrasing or word choice. Some sentences may not read as smoothly or clearly as they could in English. Despite these issues, the overall meaning is still conveyed. Please note that these ratings are subjective and may vary depending on individual preferences and interpretation.

Where are they up in arms?

How would you say your product compares to Pinecone, GCP's Matching Engine or any other product in the space?

It will be interesting to see how they define "classical" — Does it include only Western art music? Does it include only music from the Baroque and after? What about Perotin? Does it include classical musical traditions from non-western musics? Does it include folk musics? How about modern movements like 'musique concrète' or avant-garde jazz? How about instrumental film scores? Anime film scores?

Great leadership from Satya Nadella since 2014 seems to be the leading explanatory factor here. A sequence of very good decisions and products coming one after another. You forgot to mention VSCode as well which has now become an industry standard code editor. They have really focused on great developer experience.

Are we possibly becoming a post-English society in the same sense that we became post-Latin? Latin was once the language you needed to know to access knowledge and learning. Now it is of little use to most outside of the Classics. In the future it will be possible to conduct science without full literacy.

With the rise of large language models, apps, APIs, handheld computation, video content, and finger gestures, it seems that our daily interface with the world is moving from logocentrism. Furthermore, with the coming abundance of generated junk text from generative models, the relative value of language (English) looks to continue to diminish.

Not to mention __________ ______.