Sid

everything beautiful should be inconvenient and intentional

i’m so excited for it all

it's definitely not the first time I've heard similar advice provided, but I think it's still so undervalued in the current context of work that people produce. in the hype cycle of AI "researchers", even today, are just listening to all the noise and any modicum of creativity they think they have are ideas just being generated by the system that they're trying to do research on

@untethered_sid well it's also all vapid bullshit , a collection of obvious standard productivity tips u could find in a fuckin buzzfeed article. but i know that's exactly the kinda garbage that people here slurp up so

every article this kid has posted detects as fully ai generated. he claims to have been accepted to both the anthropic fellowship program for the summer and MATS for september but his posts about them do not line up with actual acceptance dates. 95% larp.

Sleep well, NY.

@itsreallyvivek @karpathy dm him in slack broski

@nik1t7n @itsreallyvivek agreed!

@untethered_sid @itsreallyvivek agree but my thinking was just to give the agent the same mindset i got from reading the article - to align my vision with the agent’s vision. because in the future it’s going to assist me in any type of research, and it must understand the paradigm i’m using

I turned @itsreallyvivek's “how to be good at research” essay into an agent skill. research-craft helps agents plan better research loops: choose problems, forecast experiments, keep logs, inspect failures, and tighten iteration. npx -y skills add nik1t7n/research-craft-skill --all github.com/nik1t7n/resear…

@itsreallyvivek @nik1t7n yea my worry as well

@untethered_sid @nik1t7n the skill is great for people who already have the mindset and just need the scaffolding but i worry it gets used as a substitute for the thing it's supposed to support.

@itsreallyvivek @nik1t7n after reading what you wrote(was very well articulated btw) i felt that the “skill” is truly an innate mindset problem. regardless of the acceleration of the process with an agent skill there is a pre-req to that, that starts with the person doing the research

@untethered_sid @nik1t7n ?

hard not to feel inspired today, so i made my own New Yorker cover.

I just want to be successful thats all..

i’m New York raised iced and praised my .99 price still alive KNICKS IN FIVE

Dude just showed up handing out pizzas, people cheering each other on to close down the street, everyone scooching over to make room for one more … never seen more love and unity. #knicksinfive

i think the biggest difference we have seen with the frontier models being produced is the step-function change in retrieval workflows which shows up the most across agentic engineering work. this + ability to still work in a meaningful manner when operating with a lack of instruction/and ambiguity has been the biggest areas of change.

@yacineMTB maybe it is the optimist in me but i genuinely believe they didn't add this constraint out of a form of controlling the competition. if that really was the case why mention it in the first place and open themselves up to this litigation?

@untethered_sid They've shown their true colors.. how do we know the same isn't done for opus..?

Wrong They are going to keep on lying to you Isn't fable actually just lobotomized mythos? So now fable is dead and you get mythos with guardrails? How could anyone trust them after this. Liars

As believers of open research, we are disappointed to see Anthropic silently degrading Fable 5 for AI development "Any topic related to building pretraining pipelines, distributed training infrastructure, or ML accelerator design... may have limited effectiveness through Claude via methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning." Not only do they get to decide what you use LLMs for in research, but this also enables them to silently intervene in your research without you knowing. This sets a dangerous precedent. If a model refuses openly, users can understand the boundary. If a model falls back to another model, users can still evaluate the difference. But if a model silently modifies or weakens its own answers while still pretending to help, researchers lose the ability to know whether a failed result came from their own idea, their implementation, or an invisible intervention by the model provider. That is not safety. Safety policies should be transparent, auditable, and user-visible. On top of that, the people most harmed by this are not the largest labs with massive teams and proprietary infrastructure. It is the independent researchers, academic groups, startups, and open-source builders who rely on public tools to compete, innovate, and pioneer AI for everyone else.

noticed something with claude generating markdown artifacts: it bakes your directional prompt into the file. tell it "build an index capturing x, y, z" and the file opens with "this index captures x, y, z" — framing that should've been response tokens gets embedded in the artifact instead. solvable with better prompting. but the interesting part is what it reveals: how much does the model actually adapt its behavior to the format it's producing? does it "know" a markdown artifact is the deliverable vs conversational output — and what does that distinction mean to it internally?