FutureTalk

#futuretalk, we talking bout the future

[FutureTalk @futuretalkin]

@karpathy Agreed, the question that begs, is how to find the sweet spot for refinement that can lead to eureka moments that are simply a multifaceted alignment/comprehension jump well beyond what exist in current RL.

[FutureTalk @futuretalkin]

@0xca0a Was just mocking up this idea last week. Of course, as always, you have the full proof in the glass puddin ready to go.

[Andrej Karpathy @karpathy]

+1 for "context engineering" over "prompt engineering".

People associate prompts with short task descriptions you'd give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window with just the right information for the next step. Science because doing this right involves task descriptions and explanations, few shot examples, RAG, related (possibly multimodal) data, tools, state and history, compacting... Too little or of the wrong form and the LLM doesn't have the right context for optimal performance. Too much or too irrelevant and the LLM costs might go up and performance might come down. Doing this well is highly non-trivial. And art because of the guiding intuition around LLM psychology of people spirits.

On top of context engineering itself, an LLM app has to:
- break up problems just right into control flows
- pack the context windows just right
- dispatch calls to LLMs of the right kind and capability
- handle generation-verification UIUX flows
- a lot more - guardrails, security, evals, parallelism, prefetching, ...

So context engineering is just one small piece of an emerging thick layer of non-trivial software that coordinates individual LLM calls (and a lot more) into full LLM apps. The term "ChatGPT wrapper" is tired and really, really wrong.

[FutureTalk @futuretalkin]

@_andrewkemp @sama Try setting up those rules in your system prompt or if you’re in an IDE set up global rules or local project rules and make sure that the system prompt is referencing

[FutureTalk @futuretalkin]

@manishrc @AravSrinivas You can do this now. Setup a Perplexity Voice Shortcut . Then pick the shortcut in Settings > Accessibility > Vocal Shortcuts, and call it Hey Perplexity, or whatever you want.

[FutureTalk @futuretalkin]

@windsurf_ai All jokes aside, months ago, one of the first things we used windsurf for was to plan out and build a rough mockup of this idea. To truly express and understand the future of vibe coding, we need to revisit what a vibe UX and AUI would be.

[FutureTalk @futuretalkin]

@martinsit @0xca0a ooooooh

[FutureTalk @futuretalkin]

@splinetool If we upgrade our account today, can we get access?

[FutureTalk @futuretalkin]

@AndrewPrifer Great work, keep it up!

[FutureTalk @futuretalkin]

@Andersonmancini @0xca0a Outstanding work as always. So glad you made it back on X and provide such wonderful use cases and clarity on how the r3f stack has updated with TSL

[FutureTalk @futuretalkin]

https://t.co/Rc5TpTRw3S

[FutureTalk @futuretalkin]

Excited to see what Anthropic and Minstral respond with.

[Min Choi @minchoi]

The wait is over.

OpenAI just dropped o1, also known as Project Strawberry/Q*

This is new level of AI that can "think" and "reason" before responding to you.

10 wild demos:

1. Coding Video Game from a prompt
https://t.co/f60r4Q0uvy

[FutureTalk @futuretalkin]

@garrettkjohnson Unfortunetly I can't share anything right now, but hope to be able to once the legal dust settles in the next couple weeks.

[FutureTalk @futuretalkin]

@0xca0a We are excited for this~!

[Nathie @NathieVR]

Big news! Meta has just announced that they're opening up Meta Horizon OS, the operating system powering Quest, to allow others to design more headsets using their ecosystem. Lenovo, Microsoft, and ASUS are among Meta's initial partners. This move could spark a new wave of headsets focused on productivity, entertainment, gaming, exercise and more.

[FutureTalk @futuretalkin]

@ILLANGELO You are your own magician!

[FutureTalk @futuretalkin]

@_LucasRizzotto @pillowsocial Bless up for the blessings of building and sharing the knowledge!

[FutureTalk @futuretalkin]

Legendary Inspiration

[FutureTalk @futuretalkin]

Coming soon…

[Rowan Cheung @rowancheung]

Apple just published a new paper unveiling MM1, a new family of multimodal AI models — with the largest at 30B parameters.

My prediction: Siri 2.0 will be the next 'ChatGPT moment' for major AI adoption.

Going in-depth on MM1 in the newsletter tomorrow: https://t.co/Gm5KwesnA7