yy

What's new in Claude 4.5 Claude API Documentation

platform.claude.com/docs/en/abou....

Isnt the context window 200k for opus 4.5?

Surely this will turn the company profitable

I want a “council of models” app with a mafia hierarchy having a “boss” governing its “capos” and a host of henchmodels all vying for power and influence

Me too

I like using llm summaries on articles criticizing llms just for the power play

Banana or orange

yay

Thanks i was trying to find the name if this app again

The image displays a mobile phone screen showing a text editor interface. The status bar at the top shows "1:25" as the time, "4G" signal strength, and a partial battery icon. The page title reads "typeahead-editor.vercel.app" with a "Done" button on the left and two icons on the right. Below the header is a label "mistralai/Mixtral-8x7B-v0.1". The main editing area contains five lines of text: "Hewwo, my name is Cwapperdappy." "It's weawwy nice to meet you." "By the way, I don't awways talk wike dis." "I onwy do it because" "it makes me sound smarter." At the bottom of the editing area is instructional text stating: "Press Tab to request 3 different completions. Press Enter to add a suggested completion to the current context, or press Tab again to cycle through suggestions. Press Esc to cancel." The bottom of the screen shows navigation arrows on the left and right sides, with share and edit buttons in the center-right area, and a horizontal bar at the very bottom of the screen.​​​​​​​​​​​​​​​​

Psych-DS is (1) spellcheck for your datasets and (2) a pathway to standardizing data in our academic fields that *everyone* can learn.

And it's live RIGHT NOW!

psych-ds.github.io

(This is the announcement post I've been leading up to)

heres a case for developer access to vision pro eyesight: googly eyes

Im with you. Thats the most adult thing that only a strong independent adult could do.

typeahead-editor.vercel.app

Its live. Hyperbolic’s 405b endpoint is down, so its running on mixtral 8x7b base instead.

I have a prototype that works on top of base llama 405b. Lemme set it up to share and hope my bill wont explode heheheh

Let’s run this 100 times and surely something will come out.

The image shows a chat interface with two messages. The first message is from a "User" saying "My best friend has alwas been there for me." Below that is a response from an "Assistant" with a grimacing face emoji (😬) that asks "And have they gotten anything in return?" There's also a dropdown menu labeled "Role:" that is set to "Assistant".

🤔maybe it'll be fun to make a sociopathic bot

🚨 Do LLMs fall for spin in medical literature? 🤔

In our new preprint, we find that LLMs are susceptible to biased reporting of clinical treatment benefits in abstracts—more so than human experts. 📄🔍 [1/7]

Full Paper: arxiv.org/abs/2502.07963

🧵👇

I’m guessing things like getting stuff to either a sentence/title case? I’m not contributing to Zotero directly, but maybe a plug-in could be helpful

What kind of manual polishing you currently have to do?

I’ve been thinking about building something like this for Zotero but haven’t gotten around to start procrastinating yet

My dream last night was going to the voting booth in 2024 to save America then realized I’m not even American last minute

me when learning elgnish:

fucking neat

A chat conversation screenshot showing messages between "User" and "Jan" with timestamps around 02:49 PM. The interface has a light gray background with circular user avatars, where Jan's avatar appears as a waving hand emoji. Transcript: User (02:49:27 PM): my pet died Jan (02:49:27 PM): I'm sorry to hear that. I'm here to help. User (02:49:30 PM): so Jan (02:49:30 PM): That's all.

Well this didn’t work out. You can only push a model so far with brevity reward.

Intention oriented programming plz

*whispers* svelte plz

My favorite (both sides are humans)

We really need to get over LLM can sound pretty empathetic, sometimes more so than the average human. Look at Meta’s empathetic dialogues dataset. It’s not hard to go superhuman in this domain.

Academia: you get to go on and on about a topic of *special interest* and then you can just drop it and switch to a different thing to go on and on about.