Moreno Vendra

Learning how to teach machines.

ML @getqonto

[Sasha Luccioni, PhD 🦋🌎✨🤗 @SashaMTL]

"ChatGPT seems so human because it was trained by an AI that was mimicking humans who were rating an AI that was mimicking humans who were pretending to be a better version of an AI that was trained on human writing." 💀💀💀
https://t.co/5qlf0wuzRB

[Moreno Vendra @MoroVendra]

@willmcgugan Not sure it's SOTA, but definitely worth checking out https://t.co/x4NZ2qlEQR

[MMitchell @mmitchell_ai]

Aah!

[Colaboratory @GoogleColab]

🔩 We've published Colab's Docker runtime image! You can connect to Colab from any machine via Docker and our local runtime connection flow. Follow the instructions below for the connection:

https://t.co/WkI2u14Vv8

[Moreno Vendra @MoroVendra]

@alexiospanos Ah nice! I actually looked into Hugo but in the end I decided to go for a custom React website + Substack for the blog! Not up yet, but planning to get everything setup in the next couple of weeks! 😁

[Moreno Vendra @MoroVendra]

Working on my personal website atm, gotta admit I missed doing a bit of frontend! Seeing the results of your work right away is so addictive! 😁

[MMitchell @mmitchell_ai]

Factuality+language models (1/2):
Large language models have evolved in the context of *recognizing* text; they are trained to recognize any text string that plausibly *could have been written*. By flipping around the "recognizing" to "generating", we can "chat" with LLMs. 🧵

[Moreno Vendra @MoroVendra]

@roydanroy Mamma mia 🙈

[Tim Hwang @timhwang]

🤷‍♂️ the great irony about this is that the tech chosen as the remedy - LLMs - is out-of-the-box orthogonal to everything search has traditionally been about: facts!

this is producing a headlong industry rush to try to bolt on factuality as a post hoc (!) feature

[Moreno Vendra @MoroVendra]

@mikispag IMO though, AI assistants are well suited for use-cases in which factuality is not a concern, given that it cannot be guaranteed. For medical diagnosis, a doctor would need to fact-check any suggestion given by the assistant.

[Moreno Vendra @MoroVendra]

@mikispag I 100% see your point, medical malpractice is a reality today, and having better tools to access knowledge and improve GPs diagnosis accuracy would be great!

[Moreno Vendra @MoroVendra]

+1

[elvis @omarsar0]

If you are looking for material to study topics related to NLP with deep learning, I highly recommend the CS224N suggested reading list.

I've read most of the items on the list at least once and they cover a good amount of relevant topics you should be aware of today.

As you start out, one of the big challenges is to pick out good material from a large amount of content available online. This is why I love the suggested readings from this course.

https://t.co/qCOBwNi2Lj

[Moreno Vendra @MoroVendra]

Yeah sure #gpt4, #LLaMA and #Claude are cool but XGBoost is still king on tabular data 👑

[Moreno Vendra @MoroVendra]

Is https://t.co/z8b5BfQxz9 down? 🫣

[Moreno Vendra @MoroVendra]

The new #ChatGPT API is officially out!

The announcement also covers the release of APIs for the underlying GPT model used by ChatGPT (gpt-3.5-turbo) as well as their speech-to-text model called #whisper.

https://t.co/2UVza0hCaC

[TrueLayer @TrueLayer]

Behind the #data👨‍💻

Learn about #machinelearning at TrueLayer and how we use @datalore for secure #collaboration, with Senior Machine Learning Engineer @MoroVendra.

Read more via @jetbrains: https://t.co/LQbno6YGB2 #openbanking

[Moreno Vendra @MoroVendra]

@minimaxir Twitter giveth, and Twitter taketh away

[Moreno Vendra @MoroVendra]

@willmcgugan I didn't know you were into ML 😂

[Moreno Vendra @MoroVendra]

@tunguz "No free lunch" applies to environment setup as well, there's no silver bullet