Lucas

Applied AI @anthropicai | prev @klarna

[Lucas @LucasSmedley]

@simonw @TheShmanuel That seems strange. Feel free to share some prompts (dms open) and I’ll try repro and get back to you

[Lucas @LucasSmedley]

@simonw @TheShmanuel Lucas from Anthropic here. On Claude Opus 4.7, display defaults to "omitted" (helps TTFT). Instead you must set display: "summarized" explicitly to receive summarized thinking. You can find more in Adaptive Thinking guide, ‘working with thinking blocks’ section

[Lucas @LucasSmedley]

@yimbyalliance @MayorofLondon

[Claude @claudeai]

Introducing Claude Opus 4.6. Our smartest model got an upgrade.

Opus 4.6 plans more carefully, sustains agentic tasks for longer, operates reliably in massive codebases, and catches its own mistakes.

It’s also our first Opus-class model with 1M token context in beta.

[Lucas @LucasSmedley]

@fabianarbor @justinmfarrugia 👀

[Lucas @LucasSmedley]

@parcadei @bcherny

[Lucas @LucasSmedley]

@amasad Some feedback: I got a replit core subscription and ran out of credits in 48hours. how is this possible - definitely didn’t use anywhere close to $25 worth of tokens on that time! I like the product but this makes me want to churn

[Lucas @LucasSmedley]

@AdeyemiJapheth 🫡

[Lucas @LucasSmedley]

@mitsuhiko A feeble attempt to lock people into the OpenAI ecosystem and make it harder to switch models to other companies

gpt-5 is around 50% higher latency on chat completions vs responses API too 😭

[Lucas @LucasSmedley]

@justinmfarrugia @frank_ Super cool! Would love if it was possible to suggest products to the site so you could crowd source ideas too

[Lucas @LucasSmedley]

@karpathy Pls remember your sample is likely high % of AI early adopters and power users

[Lucas @LucasSmedley]

@paraga @WilliamBryk @ExaAILabs my experience of Exa is better than this bench suggests, have you ran it yourself?

[Lucas @LucasSmedley]

@SimiStern @KaranVaidya6 @cursor_ai @windsurf_ai @AnthropicAI @AskPerplexity

[Lucas @LucasSmedley]

@willcb @stripepress pls but as a compilation of the seminal AI papers

[Sebastian Siemiatkowski @klarnaseb]

A lot of the reporting on AI focusses on efficiencies and cost savings. And we absolutely see lots of that at Klarna.

But our internal employee AI-Assistant, Kiki, is driving an even bigger change: helping us solve for organisational memory.

Klarna is an EXTREMELY dynamic place to work. Every month, we see 1,500 small org changes with ppl switching teams, locations and exploring new opportunities within the company. It's fundamental to our agility.

But it makes it even more important that we centralise the knowledge and experience which each individual holds in their heads, so we can build on that even when they have moved to a new team.

Kiki is how we do that. It's like being able to ping anyone in the entire company and get an answer immediately.

https://t.co/oObJnB7eNy

[Lucas @LucasSmedley]

@krea_ai It’s just timing out at the minute for me 😢

[Lucas @LucasSmedley]

@NaveenGRao @MosaicML @databricks @jefrankle Hi, need to be verified to DM you but please can you drop me a message. Interested!

[Lucas @LucasSmedley]

@jxnlco The tutorial on knowledge graphs is significantly improved and easier to follow now 👍 my only feedback is the real world example could be a bit more based on an actual problem someone might be aiming to solve

[Lucas @LucasSmedley]

@jxnlco @ivanleomk Hi, do you have any notebooks available on this? Or will we be able to see through Instructor at some point? Thanks :)

[Lucas @LucasSmedley]

@jxnlco If you wanted to extract nodes and edges based on a given pre-existing ontology how would you think about that? Let’s say you already had a graph which had a certain schema you wanted to follow