Keep on to blur preview images; turn off to show them clearly
In both prose & computer code, IMO the rule for current-generation AI (even SOTA) is simple: Don't use it for thinking, and don't even use it for writing -- use it for typing. The more of the cognitive load you keep for yourself & just let the bot type, the better the output.
Author. Coder. CTO. θηριομάχης. Building: https://t.co/otXT4Wy6WR. Writing: https://t.co/dBPBtyCIHw.
This is probably a place where the AI has mangled the details, & he only skimmed the output but didn't flag this. This exact same thing happens with coding models all the time & causes bugs. Wild to see the same thing in published work.
In both prose & computer code, IMO the rule for current-generation AI (even SOTA) is simple: Don't use it for thinking, and don't even use it for writing -- use it for typing. The more of the cognitive load you keep for yourself & just let the bot type, the better the output.
RT @0xSigil: First time in Europe! Honored to speak with @GEVS94 from @a16z, @gorkem founder of @fal, and @eliast founder of @agency - sh…
Investor @a16z | God chose the foolish things of the world to shame the wise - 1 Cor. 1:27 | 🇸🇻
building an agent harness is an exercise in distilling your team’s opinions + knowledge into your product. some examples: - when to do context compaction, Claude Code doesn’t want you to get to the 200k limit bc quality degrades for users much before. this is an informed opinion - what granular tools get rolled up into bigger semantically grouped tools - what tools to use! for example, not all web search is created equal - what subagents to create for context management, specialized instructions, etc. should subagents use different models - how to facilitate planning via prompts and hooks - how to summarize, eject, and organize tool calls. an option is to store them in a filesystem - bring docs locally, are they small enough to load everything in or do progressive disclosure via skills you don’t have to do all of this, but it’s a great exercise to organize, debate, and gather data on what your agent behavior should like and how you know your decisions will yield that behavior. debate, dogfooding, and evals are your friend
building agents and harnesses, prev @awscloud, phd cs @ temple
Most underrated warlord in history and one of the most Based political leaders ever. On reflection, crushes competition like Ramesses II eclipsed Alexander for Adrian Veidt. Not even @SamoBurja gets it. Should be the patron saint of everyone who cares about good governance.
We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
RT @HamelHusain: Next week @sh_reya is hosting a unique session on processing unstructured data at scale with LLMs. This session is the cu…
Asst professor @MIT EECS & CSAIL (@nlp_mit). Author of https://t.co/VgyLxl0oa1 and https://t.co/ZZaSzaRaZ7 (@DSPyOSS). Prev: CS PhD @StanfordNLP. Research @Databricks.