dis

he/him

[Cory Doctorow NO LONGER ON TWIT TER @doctorow]

AI slop turned Facebook into an anaerobic lagoon of botshit, the laziest, grossest engagement bait, much of it from rise-and-grind spammers who consume get rich quick "courses" and churn out a torrent of "shrimp Jesus" and fake chainsaw sculptures:

https://t.co/NxsYuVfDcm

16/

[dis @Seferin4]

@johnlashlee @zlflynn Batch processing data where it is stored is probably a more cost effective approach in most cases but some problem spaces are too complex to pre-compute and are best solved at runtime. Also building runtime services with plumber is pretty good fun :)

[dis @Seferin4]

@predict_addict Financial gradient descent?

https://t.co/uLhj6aiC1B.

[dis @Seferin4]

@Swayam_Nath_ I suspect you're just not getting good stuff. Check which side of the cold war the intellectual tradition is on.

[dis @Seferin4]

@martin_eve In the UK corruption generally seems to take place in public, in accordance with the law, without much recognition and in the near absence of the widespread moral scandal it so richly deserves.

[dis @Seferin4]

@pressfuturist I don't think work as personal as a thesis can ever be perfected or even completed to satisfaction so it can never be finished. The only thing you can do is finalise it to be abandoned so that the great relief of working on something else can happen. Good luck!

[Valeriy M., PhD, MBA, CQF @predict_addict]

https://t.co/9qBErXWLeB

[Philipp Schmid @_philschmid]

A small, portable vector database powered by SQLite for on-device RAG? 🤯 sqlite-vec is a new vector search SQLite extension written entirely in C with no dependencies, MIT/Apache-2.0 dual licensed.  

sqlite-vec queries:
- 1 million 128-dimensional vectors in just 17ms
- 500,000 with 960-dimensional vectors in 41ms

sqlite-vec supports:
💾 Matryoshka embedding slicing
💡 Binary quantization reduces storage by 32x with minimal accuracy loss
🤏🏻 L2, cosine and Hamming distance calculations
🧮 Retrieval against Python List and NP Arrays
🛠️ SDKs for Python, Javascript, Go, Rust, Wasm and more
🧠 local direct embedding with “sqlite-lembed” for gguf models and @ggerganov Llama.CPP
☁️ remote embedding with “sqlite-rembed” for @OpenAI compatible APIs

[dis @Seferin4]

@xgabegottliebx In the aggrandised socio-historical-linguistic sense one would have to make it impossible to write in some important way as everyone had written before. Maybe you have written a particularly decisive paper?

[dis @Seferin4]

@xgabegottliebx One could almost think about deconstruction as a transcendental argument in reverse.

[dis @Seferin4]

@xgabegottliebx If we deflate (and risk psychologising) the term something like deconstruction must occur whenever we change our mind.

[dis @Seferin4]

@xgabegottliebx Instead of apprehending the general necessary preconditions of a fact; a general necessity is revealed as an impossible assumption in the confrontation of the fact.

[Valeriy M., PhD, MBA, CQF @predict_addict]

The paper we have been waiting for essentially shows that
#timeseries #llms do not work in forecasting.

Back in 2022, paper “Are Transformers Effective for Time Series Forecasting?“ challenged the appearing narrative that transformers are useful for forecasting. By removing transformer elements the authors showed the performance went up ⬆️

And now people did the same with time series LLMs. The papers demonstrated:

- removing the LLM component or replacing it with a basic attention layer does not degrade the forecasting results—in most cases the results even improved!

- in fact removing even removing the language model entirely, yields comparable or better performance!

- these simpler methods after removal of LLM component reduce training and inference time by up to three orders of magnitude while maintaining comparable performance!

- the sequence modeling capabilities of LLMs do not transfer to time series. By shuffling input time series the authors find no appreciable change in performance.

What this says is that LLMs can’t deal with critical features of time series, the time order is key and if LLMs performance doesn’t change when shuffling data it basically means it doesn’t model time series.

These finding are as damming to time series LLMs as the “Are Transformers Effective for Time Series Forecasting?” was for transformers.

#timeseries
#forecasting

[dis @Seferin4]

Don't worry, all politics is human geography and all human geography is history and all history is phenomenology.

[Anya Martin @AnyaM8_]

Interesting - think this is the first time I've heard someone critique "omnicause" campaigning outside of very online policy circles. Cutting through

[dis @Seferin4]

@matthewclifford @HenryEOliver This is some ignorant shit. GP in your pocket? Please. Medical advice is high risk under the EU AI Act and SHOULD BE because guessing about medical advice will kill people and that's what AI means - guessing. I would not take advice on what over regulation means from this man.

[Pankaj @pankajk_21]

@soychotic RStudio is the asylum that raised me

[dis @Seferin4]

Stand back advertisers - someone just said something sensible about #AI on Twitter!

[dis @Seferin4]

This interesting thread swerves into eugenics at the end - the explanatory variable here is going to be culture not genes.

[Alex @notcomplex_]

The Chinese Communist Revolution consolidated its power by taking land from the elite. In the following generation, the previous elite once again reclaimed their social advantage.

This is not an isolated phenomenon: A Thread on the persistence of status 🧵

[dis @Seferin4]

@kareem_carr You can fit an LM via OLS or gradient descent. The resulting model weights and summary statistics can be equivalent but the closed solution is not ML and the latter is. Despite all the hype and bullshit there is a shift in methodology there; though the future mightn't notice.