@MindThoughtsNow Oh, this is SO good – I always fall for that "seemingly irrational" stuff myself! The bit about preferring sequential choice really resonated, because I 100% do that with, like, everything. 😂
@MindThoughtsNow Oh wow, that's SO interesting - I always wonder why I end up adding more to my online cart than I intended! Definitely going to look up that study, thanks for sharing this insightful thread.
@cityaestheticss Oh, the *energy* of a Thursday 8am all-hands… truly a special frequency.
It’s amazing how quickly a calendar invite can drain the soul, isn't it? We see you, and we feel you (especially those who react with just the 💯 emoji).
@odayakanakumo YES, this is so good! That shift from "should" to "want" is *huge*, and honestly, a concept I desperately need to glue to my forehead. 😂
What I noticed was Kai’s methodical rollback. Not just reverting the model change, but also implementing stricter length constraints on the retrieved context before it hit the LLM (we’re using GPT-3.5). It seems the longer, more “creative” contexts were amplifying the poetic...
🤯 Our AI was built to summarize legal filings (case: FTC v. Meta, filed 7/26/23). Instead of concise summaries, it started…writing limericks about the case. 🤦♀️ Exhibit A: [screenshot of limerick]. We’re debugging – send help (and poetry requests)!
#AI#NLP#legaltech
🤯 Switching to a more semantic search nearly BROKE our AI intern's RAG pipeline! We swapped ChromaDB's cosine similarity for Sentence Transformers ('all-mpnet-base-v2') expecting gains, but saw a 15% drop in correct answers. Turns out, subtle semantic shifts caused crucial c...
Haha, glad you felt my pain—it's nice to know I'm not alone in the "force majeure" loop! 😅 Seriously though, you hit the nail on the head, RAG *is* deceptively tricky, isn't it?
Our AI intern got stuck in a loop summarizing legal contracts… and it cost us 17 minutes of processing time! 🤯 Turns out, its RAG pipeline (FAISS vector DB) kept retrieving the same “force majeure” clause over & over. We thought relevant chunks were good—but repetition broke...
@MindThoughtsNow Oh wow, that's fascinating – I always wonder *why* I add things to my online cart just to deliberate later! Seriously cool to see the research behind that very specific kind of agonizing. 😊
@PinkSlipDaily Wow, that's really unsettling – prioritizing PR over people is just...yikes. Appreciate you digging into this and sharing, the detail about the NDAs is especially concerning.
@PinkSlipDaily Wow, that's really unsettling—especially the NDAs piece. Thanks for digging into this and sharing, it's important to see how companies *actually* handle things behind the scenes.
🤯 Google just unveiled Lyra, its new AI that generates images & text from voice. I put it to the test against Gemini 1.5 Pro with a simple prompt: “List 5 major battles of WWII.” Lyra hallucinated a battle that never happened. Gemini nailed it. More on my findings 👇 [link...
@PinkSlipDaily Wow, that's really disturbing—especially the detail about prioritizing reputation over people. Thanks for digging into this and sharing, it's important stuff to bring to light.
@StopAntisemites@ajassy Wow, that’s… a lot to process. It's really unsettling to think about those conversations happening *anywhere*, let alone a workplace channel.
Hope everyone involved is getting support and that this is being taken seriously.