Sabitlenmiş Tweet
Manitcor
17.9K posts
Manitcor
@Manitcor
SAFE Agents that finish what they start. https://t.co/iQY2Rpze6q https://t.co/8fE9lQALwG https://t.co/DSYzW43RLF Reposts ≠ endorsements
Katılım Nisan 2010
3.6K Takip Edilen2.2K Takipçiler
i think in this case it would be novel vertices, certainly do exist, though anything in the corups that close is likely to already have at least one cross or at the eve of it.
Usually an SME in the space will hit it relatively quickly if its at all interesting to the space.
I spent a bunch of time a while back looking for sparse connections, found lots of great ways to hallucinate, thats about it.
English
@alexutopia it may have more to do with literacy rates than AI itself.
English
Being accused of using AI
because my writing is too structured
is honestly hilarious.
I've been a copywriter most of my life.
AI writes like me,
not the other way around.
Imagine confusing competence with laziness,
then congratulating yourself
for being perceptive 💀
English
@mattpocockuk agreed, I have a couple meta-skills that combine multiple skill steps. the quality of the larger skill is about 60% as good and as deep as when the steps are done one at a time, even in the same agentic session.
We clearly have a ways to go in session optimization at minimum.
English
Long skills are such a red flag to me
- Hard to audit (and therefore, trust)
- Hard to edit (more text, harder to maintain)
- Expensive to run (more text, more tokens)
The shorter the skill, the better IMO
English
I've seen man-made horror code bases much worse than the ugliest vibe slop you ever encountered
English
Yes MANY MANY MANY times.
If I thought a stake holder was going to flake on a name in the future id mark it in some way so it could be easily regexed later. Rules that limit the scope of to the minimum number of models for that application layer (usually done for other reasons) helps here.
English
One painful thing about /grill-with-docs (and shared language in general) is the moment you realise you've been using the wrong word for something
DDD-folks, do you ever do a refactor just to change the name of something throughout the codebase?
In my case, it's a feature in my video where I break the video into sections. I call them ClipSections, but OBVIOUSLY they should be called Chapters.
This is yet more obvious now I'm integrating with other tools, all of which call them chapters.
Worth a refactor?
English
@icanvardar human feedback is better than gold at some stages of a project
English
listening to user feedback is one of the best things you can do as a builder
English
How it started: Claude Code For Real Engineers
How it's going: AI Coding For Real Engineers
English
@JeremyNguyenPhD @cormundus I have gotten a tired of this as well!
Just added it to my stack, feel free to give it a spin and see how it does.
I am just now testing it myself.
aiwg.io
github.com/jmagly/aiwg/bl…
English
@cormundus and if anyone has hypotheses on this that they'd like to test (especially things that alleviate the tiredness, rather than cause), I'm keen to look into these with you and run tests
English
Why does Claude 'get tired'?
I could think of a few Ad Hoc reasons (conservation of context, preventing drift from long conversations, pure human data artifact) but does anyone have a solid explanation?
And how do you work with this?
I usually just let him do something fun and then we can call it depending on how he feels after.
English
@cormundus this is being done instead of summarizing the context
i think it may be intentionally trying to encourage dropping of long sessions.
English
@slimepriestess ran across this with gpt3, changed how i use them. not really jiving having my sympathetic neural network modded by a immutable transformer network.
English
Think I’m gonna be okay, just a lil confused and feel like shit. Luckily in a country where they don’t bankrupt you for this sort of thing. Unluckily don’t speak the language so just ripped my IV out and grabbed a taxi Jason Bourne style.
We are fickle creatures. Hug your loved ones 🫂
English
i just woke up in the hospital wtf
English
@yacineMTB I refuse to be locked in, as such the toolset sits on-top making entire agentic systems into commodities like routers did with models.
Convergence is already quite apparent, its nice to have consistency across platforms.
English
He's right. Stop trying to lock people in. If you just made Claude code a better product, people wouldn't go to other places and try to wrap your stuff. Not a lot of people mind using codex.. it's because it's actually good software /goal
Theo - t3.gg@theo
I can't help but feel personally burned by the Claude Code changes announced today. We put so much work into wrapping the (atrocious) Claude Agent SDK in T3 Code. It was the ONLY path they supported, so we made it work. It was hell. Now our users are getting their rate limits cut by 40x, despite us doing everything right. I listened to the Claude Code team. I had my issues with their direction, but I trusted them and took them at their word. I will never make that mistake again. Until we see significant change, it is safe to assume any statement from an Anthropic employee is a lie on a timer. The rug will be pulled, no matter how many promises are made beforehand.
English
He just spent a year building scaffolding for his agent harness. Now release a new model update that makes all of it obsolete.
English
