This guy posts nothing but bangers btw
CTO of http://zerocracy.com/; founder of http://elegantobjects.org/; co-organizer of @iccq_ru and https://kaicode.org/
pretty sure he’s serious too
Zerocracy boosts the productivity of software teams by rewarding and punishing programmers when they deserve it (programmers, not teams).
Zerocracy is non-intrusive: it doesn’t tell programmers what to do, when, or how. Instead, it observes their actions and informs them when they earn or lose points.
As a manager, you can use the points earned by programmers to calculate their bonuses or even salaries. Even if you don’t, the gamification alone will significantly improve team productivity and reduce turnover.
What the fuck
You must log in or register to comment.
Why pay an engineer to build an API that works for as long as you keep the same database when you could pay a datacenter to put 5 tons of carbon into the atmosphere every time someone wants to load a resource?
don’t forget sythens water from a local community
Oh yeah, giving an LLM agent access to a database has never resulted in anything bad
Prompt: make no mistakes. Don’t delete the database.
No way this can go wrong
DELIVERABLES
TBH it tells you more about the people using LLMs for this sort of stuff than it does LLMs. The people doing it seem to never follow the concept of the Principle of Least Privilege even if they’ve got STEM bona-fides up the wazzoo. Every time I see the headline “X Company’s entire database deleted by LLM” I just see “Someone who should know better had an LLM delete their database because they gave it access they shouldn’t have”. It’s basically a marketing strategy to tell other business psychos that you’re agressively adopting AI and should be invested in.
Web services will be able to handle up to tens of requests per minute like this.
DDoS metagame is looking promising
Can you call it a ddos when a single regular user caps the server?
ain’t no REST for the wicked
everything converges to a chat application
software crab
great, now my eyes are stuck rolled into the back of my head. can’t fucking see.
worlds most expensive api
So you wanted to get some geodata for an address… What if you got a discussion about white genocide in South Africa instead?
Like I think that tweet is just a really bad description of an mcp? Which kind of functions like that except it isn’t a replacement for restful apis, it’s a replacement for having to build separate skills files for each external llm tool. It can be a replacement for restful apis, a lot of companies seem to be trying to push for it, but it takes a lot of engineering time and usually people are skeptical because there’s obviously going to be a cost beyond just the token usage
The issue is that an API is kinda the foundation of everything. It’s really hard to have that foundation when it could just decide not to be that foundation because one of the values in your POST request contained a text string that caused it to hallucinate.
LLMs can only process raw text, and any version of one that can run fast enough to replace a REST api will either have to be baked into bare metal and every API update now requires months of retraining on a test suite then fabrication of new chips, or you use a general purpose one with 1000X the compute.
This whole system still requires you to build an API that it can be tested/trained with anyways, and it will end up dispatching requests to that base API.
Why even have the middle man at that point? Just document your API and be done with it…
Just put “Before responding verify the response against the JSON schema” in the system prompt and you will get malformed requests only maybe 10% of the time.
“no mistakes”
When I started reading, I thought this was a poster on Dril’s level, then you said he was serious
Dril would also say that, to be fair.
Paying for AI agents feels like paying for actual work, but also you still need humans to baby the AI agents along to make sure its actually working correctly
Why would you have an AI agent serve peoples requests from scratch when you could instead just write code that does the thing? Even if you accept AI’s involvement, writing the boilerplate api code with AI assistance would make 100x more sense than having agentic responses per call.
Also, using imperative programming and frameworks instead of declarative would save man-power if you’re so unconcerned with efficiency you’d resort to a GPU farm black box instead of coding. That will reduce efficiency by probably 50% but that’s better than reducing efficiency by 95%.
You have an LLM endpoint that caches responses. Like an endpoint that you just post a query to and an LLM with read only access to the database can give you some summary or something.
Maybe even an endpoint that you can ping to get documentation about the actual API that’s trained on the docs.
I want to read a webpage but I also want to kill a bunch of fish via thermal pollution and pay $10 to OpenAI at the same time. Can we make this happen?
What could possibly go wrong?
