tim_sw

False. Eric Schmidt is one of the sharper operators in tech and his effect can be seen when Google started declining after 2011 when he stepped down.

My 2 cents - don’t rely on these frameworks and just do it yourself (or pick libraries like Instructor over these frameworks)

I think both have the wrong abstractions for people to build more complex workflows and use cases beyond demos.

Pastebin https://pastebin.com/StEkp0YT

How does this compare with HyperDX?

https://news.ycombinator.com/item?id=37558357

This is more common than reported. I’ve lost a semi expensive present, and have heard anecdotal stories of people losing handbags, jewelry, higher end clothing/shoes.

It seems to be more prevalent in the US/TSA than other first world countries.

This is a very high quality writeup.

this is not a duplicate - it's not the announcement post, it's a deep dive into the paper from the POV of Hugging Face's RL lead.

Defense against tiktok and bytedance

The blog’s name is “Software The Hard Way” and this post lives up to that name

The devil is in the details. Training large LLMs requires a lot of custom infra (handling GPUs going down, efficiently pushing data to keep the accelators busy, deciding on which mechanism of parallelizing model training is better - data vs model parallelism or both, tuning hyperparams of optimizers which can be different for larger batch sizes, etc)

Mosaic is one of the better providers for this. AWS is nowhere near ready at this current point in time, it is pretty much a "dumb" infra provider in large LLM training at this point. (Of course they won't be standing still and will prob acquire that capability one way or another)

While somewhat useful - there are no systematic ablation and comparison studies, datasets, and quantitative evals here. It's all anecdotal, so should probably be a blogpost and not a "paper"

We find that responses from existing generative search engines are fluent and appear informative, but frequently contain unsupported statements and inaccurate citations: on average, a mere 51.5% of generated sentences are fully supported by citations and only 74.5% of citations support their associated sentence. We believe that these results are concerningly low for systems that may serve as a primary tool for information-seeking users, especially given their facade of trustworthiness. We hope that our results further motivate the development of trustworthy generative search engines and help researchers and users better understand the shortcomings of existing commercial systems.

this looks like it's for GCP. TPUs are used for most internal workloads. It's available externally but some of the papercuts and devex without the TPU/TF team helping you can be more painful than using Nvidia/CUDA

Pretty high model flop utilization:

—————

MaxText is a high performance, arbitrarily scalable, open-source, simple, easily forkable, well-tested, batteries included LLM written in pure Python/Jax and targeting Google Cloud TPUs. MaxText typically achieves 55% to 60% model-flop utilization and scales from single host to very large clusters while staying simple and "optimization-free" thanks to the power of Jax and the XLA compiler.

How are people using vector DBs in production? Do you typically use and manage Faiss indexes alone or use something like Milvus, Pinecone, Weaviate, or Chroma?

Fun tweet from @amasad

https://twitter.com/amasad/status/1640831001256689664?s=46&t...