And I'm betting that there's an inverse token consumption pyramid showing those 5,000 people consuming, by far the most tokens, and the 1 to 2 billion people consuming far fewer. And I'm not talking just on a per capita basis.
Or if the consumption pyramid doesn't look that way right now, it will very soon.
Fascinating to think that this was impossible to write or even think about 4+ years ago: openai.com/index/an-alien-mi…
What an amazing journey Groq is on. From creating a blazing fast inference chip to building the world's premier, at scale inference cloud
We're just getting started. Thanks for your partnership, @nvidia & @Dell
The next wave of AI is about deploying incredible innovation at massive scale.
Excited to work with @GroqInc and @nvidia to make that happen with @Dell. 🚀
xitter.cf/GroqInc/status/2091908…
MCP is finally & appropriately going stateless blog.modelcontextprotocol.io…
The "Price-per-task vs price-per-token" section Is an example of why obsessively focusing on token cost is idiotic. You have to measure models at a higher level of abstraction to really understand their performance as well as their clock speed. It's kind of like measuring a vehicle exclusively on an MPG basis: both a Ferrari and Escalade have the same MPG, but one moves more people per gallon of gas used. P
databricks.com/blog/benchmar…
My dad grew up in Romania, under one of the world's most oppressive communist regimes. When I was a kid, he would always say "America is the best country on Earth". He wasn't wrong.
Thanks for being America, America. Happy 250th
For those of us with hyperactive inner monologues, we can totally understand how an LLM-based conversation protocol can control action and activity in the real world: reddit.com/r/Damnthatsintere…
Happy Thanksgiving, all. I'm thankful for the tremendous amount of Black Friday spam and how today serves as the ultimate 'unsubscribe' reminder.
Warning, this is a mouthful: AI is the wholesale outsourcing of the crowdsourcing of anything crowdsourcable from a limited population with limited capacity for cognitive work (people) to an unlimited population with unlimited capacity for cognitive work (agents)
Sinclair Schuller retweeted
this is strangely heartwarming: the canadian pediatrics association now recommends that children engage in risky play—"thrilling and exciting forms of free play that involve uncertainty of outcome and a possibility of physical injury"—because of benefits e.g. to mental health
Sinclair Schuller retweeted
Yes, your boss wants you to know how to use AI - but don't worry, you don't have to be an expert.
That's according to @LinkedIn's chief economist and other labor market pros.
New for @BusinessInsider from @mhofftweets and I:
@sinclairx @nuvalence @hyams
businessinsider.com/artifici…
I wrote a taxonomy blog post to help answer the question "what type of thing are we building?" when discussing generative AI solutions. The taxonomy aligns GenAI solutions with one of six categories.
nuvalence.io/insights/a-6-ca…
We've been spending a lot of time discussing AI with our clients and amongst ourselves. My latest post focuses on product strategy & four pragmatic categories on how to integrate business with conversational AI tech like ChatGPT: nuvalence.io/insights/4-ways…
So Putin decides to invade Ukraine because he wants to keep NATO "off his porch" in order to defend Russia, but taking Ukraine increases how many linear kilometers Russia will share with NATO countries...got it
