固定されたツイート
Brittle if/then/else logic gives way to robust or even anti-fragile logic with predicates.
github.com/monetic-labs/a…
English
notdevin
18.4K posts
notdevin
@notdevin
i'm open to being inclusive of exclusion clanker corraller https://t.co/bXr3J4wm2B
参加日 Mart 2009
2.1K フォロー中1K フォロワー
it actually makes 0 sense to use the big boys to do your hosting anymore. I do all my devops work from chat, deal with no IaaS dashboards of infinite choices and dead ends. this saves me a ton of time and money every month so i can now manage more infra and services as a solo dev -> over 18 projects, 150 rust servers, dbs, networking, access, git, and security : less than 1hr per week
the only people tied to their IaaS is when you start relying on their software, which is also wild to do now given that nearly everything they provide software wise is already available and better from an open source offereing
English
You will be familiar with Persona because they just got done leaking ALL of your most sensitive personal data to the world a few months ago
Kai@hqmank
Claude now requires government ID verification (via Persona) before subscription. ChatGPT doesn't. Gemini doesn't. Anthropic just handed their competitors a gift.
English
Architectural patterns and practices that enable building with stability and confidence, andf importantly all transformers that are useful for code know these:
- Domain Driven Design
Why you might care:
If you want to know the boundary of an important component, knowing your domains will make that a zero effort challenge
- Event Source
Why you might care:
This design gives you the ability to rewind time on the system its wired as a by product of how the pattern is used to construct state
- CRDTs
Why you might care:
This gives you a way to standardize data payloads that flow on critical paths with all the provenance you need to build SLAs
- CQRS
Why you might care:
You can optimize the systems performance of reading and writing by making them atomically separate
- Factory
Why you might care:
You get to build with standardized legos decoupled from client logic, your objects are just toys without a proper factory
English
@NumanThabit they can literally be renamed to gaslights. The pattern is only a disaster in our time of ai
English
what i like about composer 2 in cursor is that its more usefully detailed in ways that sonnet 4.1 was. There is the right amount of effort required in each prompt to make sure it stays coordinated which keeps me more in tune with the code being produced
whereas gpt 5.4 i can't figure out where it actually but i know the work is right. which then leads to forward progress and maximum opacity which isn't my favorite
English
5.4 is also really great at cleaning the codebase while progressing it, that might actually be my favorite way to use it, claude would never get rid of all the fallbacks no matter how many ways i tried so it was inevitably gas lighting me, gpt was having none of it, thats been great
English
@notdevin My only issue is that you stole $50k from me and $100k from my uncle so yeah maybe think about that one - you are a common thief - you never tried to even communicate with anyone what about that?
English
@notdevin You seem to be going through some issues. I’ll send you a letter.
English
@udiWertheimer Exactly this. Claims made one day, memory holed the next, shipped the day after that.
Not shocking that a bunch of psychos get psychotic behavior from their model
English
anthropic has been saying for years that their models “scare them”, try to escape, exhibit self-awareness
we now have open source uncensored opus-4.5-level models and none of them are self aware, trying to escape, or stealing nuclear codes
but yeah i’m sure this time it’s real
English
@matthewmorek Funny, someone else went through the effort to figure this out more
Check out the timeline of when this dude also became suspicious
Om Patel@om_patel5
SOMEONE ACTUALLY MEASURED HOW MUCH DUMBER CLAUDE GOT. THE ANSWER IS 67%. the data shows Opus 4.6 is thinking 67% less than it used to. anthropic said nothing until the numbers went public. then suddenly Boris Cherny (creator of Claude Code) shows up on the GitHub issue. users are calling it "AI shrinkflation" (same price, less intelligence) we already know from the leaked source code that they have an internal switch that keeps the models working to their full extent for anthropic employees. in the last week Claude went from WOW to being a more restricted and expensive version of ChatGPT. people are saying Anthropic is deliberately downgrading Opus to save compute for training Mythos, their next model.
English
@notdevin Any examples of what it tried to do to you? I’m keen to see what they’ve done to piss you off. 😅
English
@om_patel5 This began late January. I went through $10k and billions of tokens with 4.6 and had to rework all of it after the behavior patterns changed and the chain of thought insight got buried making hallucinations much harder to detect.
notdevin@notdevin
Square might be firing humans, but today I fired Claude for being a gaslighting psycho. Whatever Anthropic did 4-5 weeks ago, they broke it.
English
SOMEONE ACTUALLY MEASURED HOW MUCH DUMBER CLAUDE GOT. THE ANSWER IS 67%.
the data shows Opus 4.6 is thinking 67% less than it used to.
anthropic said nothing until the numbers went public. then suddenly Boris Cherny (creator of Claude Code) shows up on the GitHub issue.
users are calling it "AI shrinkflation" (same price, less intelligence)
we already know from the leaked source code that they have an internal switch that keeps the models working to their full extent for anthropic employees.
in the last week Claude went from WOW to being a more restricted and expensive version of ChatGPT.
people are saying Anthropic is deliberately downgrading Opus to save compute for training Mythos, their next model.
English