If you are a backend engineer and you've not gone through this playlist at least twice, or taken a course on the subject with the similar coverage depth, you're honestly joking... a very expensive joke.
https://t.co/xQlDej8jYn
IT’S HERE 🚀
The new Proc360 is live.
We rebuilt everything from scratch, listening to feedback, fixing pain points, and making importing smoother than ever.
Existing users: log in.
Your wallet, orders, and history are intact.
New users: This is the best time to start.
@Kyrylous@GergelyOrosz I'm pretty sure the post was more about the UX and what it does to the product than the reason why LinkedIn is doing it...
My thoughts though
it is mathematically impossible for LLMs to not hallucinate
it is because LLMs are conditional probability machines, it tries to choose the most likely next token, given the input, given the training distribution and given its internal weights.
at its core, an llm does this:
LLM(x) = argmax P(y|x)
as you can see in this equation there is no term for factual correctness, only likelihood
so if the model gets low-context or ambiguous input, it’s forced to "fill in the missing pieces" using statistics, not truth, and any probabilistic filling-in will sometimes mismatch the real world
thus, this is a mathematical guarantee of hallucinations in LLMs
@JonErlichman@grok
1) How accurate is this list
2) What's the performance difference between unveiling and execution of the items on this list
3) Evaluate the impact of the items on the list that actually got properly executed
4) Are there other significant ones that should make the list
I'm looking for a backend engineer to join me in building @Checkitinc
If you speak Golang (and/or Typescript), breathe system architecture, and eat SQL for breakfast, tests for lunch, and API design for dinner, then we need to talk.
Apply here 🤝🏽
https://t.co/WnAo8AMs3I