Here's qwen-30b-a3b's response to your prompt when I worded it better:
The prompt was:
"Create a Python decorator that registers functions as handlers for MQTT topic patterns (including + and # wildcards). Internally, use a trie to store the topic patterns and match incoming topic strings to the correct handlers. Provide an example showing how to register multiple handlers and dispatch a message to the correct one based on an incoming topic."
Are you using Ollama? If so, the issue may be Ollama's default context length: just 2,048 tokens. Ollama truncates the rest of the context silently, so "thinking" models cannot work with the default settings.
If you are using Ollama, try explicitly setting the `num_ctx` parameter in your request to something higher like 16k or 32k, and then see if you still encounter the looping. I haven't run into that behavior once with this model.
The real definition of vibe coding is coding with just an LLM. Never looking at what code it outputs, never doing manual edits. Just iterating with the LLM and always pressing approve.
It is a viable way of making software. People have made working software with it. It will likely only ever be more prevalent but might be renamed to just plain old making apps.
I think your prompt is bad. Still impressive that Claude 3.7 handled your bad prompt, but qwen3 had no problem with this prompt:
Create a Python decorator that registers functions as handlers for MQTT topic patterns (including + and # wildcards). Internally, use a trie to store the topic patterns and match incoming topic strings to the correct handlers. Provide an example showing how to register multiple handlers and dispatch a message to the correct one based on an incoming topic.
I purposefully used exactly the same thing I did with Claude and Gemini to see how the models dealt with ambiguity. It shouldn't have degraded the chain of thought to the point where it starts looping.
The trick shouldn't be to try and generate a litmus test for agentic development, it's to change your workflow to game-plan solutions and decompose problems (like you would a jira epic to stories), and THEN have it build something for you.
IMO, if you're gaining a significant amount of productivity from LLMs in a technical field, it's because you were either very junior and lacked much of the basic knowledge required of your role, or you performed like you were.
I disagree. Maybe there's savants out there that can write SQL, K8s auto scaling yaml, dockerfiles, React components, backend code, and a dozen other things. But for the rest of us LLMs are helpful for the things we wade into every so often.
It's not miraculous but I feel like it saves me a couple hours a week from not going on wild goose chases. So maybe 5% of my time.
I don't think any engineering org is going to notice 5% more output and layoff 1/20th of their engineers. I think for now most of the time saved is going back to the engineers.
Definitely not the case for coding. I'm a capable senior engineer, and I know many other very experienced senior engineers who are all benefitting immensely from AI, both in the code editor and chat interfaces.
My company just redid our landing page. It would probably have taken a decent developer two weeks to build it out. Using AI to create the initial drafts, it took two days.
IMO, if you haven’t been getting a significant productivity boost from LLMs in a technical field, it’s because you lack the basic brain plasticity to adapt to new tools, or feel so psychologically threatened by change that you act like you do.
I would (similarly insultingly) suggest that if you think this is true, you're spending time doing things more slowly that you could be doing more productively by using contemporary tools.
Nearly every job, even a lot of creative ones, require a degree of accuracy and consistency that gen AI just can't deliver. Until some major breakthrough is achieved, not many people doing real work are in danger.
I wonder what his agenda is. He's really being willfully dishonest here, like he was when he kept bringing up the 200 year-olds in the Social Security databases, without mentioning that they weren't actually receiving payments. I don't get why he needs to mislead us like this if his intentions are pure.
That's a big, massive if. I think the more experienced you are, the less net productivity you gain. If you're more senior, is probably very low or even negative. I use ChatGPT every day pretty much, but it has wasted quite a bit of my time.
When it does save me time, it's almost just some boilerplate I didn't have to Google or type out. Honestly, I feel like gen AI pleataued a year or more ago.
LLMs are just really convincing bullshit generators. They look impressive on the surface, and the times when they spit out a whole bunch of useful boilerplate feels like magic, but that stuff isn't super useful for a majority of the work you spend your time doing. Throwing more money at these companies is not magically going to yield AGI. I think the AI CEOs are basicallly selling us a lie.
I think experiences are highly variable based on the person and stack. I'd say I've gotten a ~50% productivity boost. I'm quite senior, make FAANG money.
I work in typescript, rust, and go - mostly typescript. LLM coding is an order of magnitude better at typescript than these other languages. To contrast, LLMs seem mostly useless for rust.
I also use cursor, which is a big jump over other code assistants.
And finally I understand how to prompt LLMs accurately. This is a new tool and from interacting with coworkers in the same codebase I can say many smart people have not yet learned how to use the tools effectively.
But if you just assume that everyone catches up to where I am today with cursor + typescript the change is massive.
One of my co-workers is going all in with one of them in his IDE, as part of a pilot program for using it at our company, and ever since it seems like he's gotten dumber, but by outside metrics he's probably "more productive": Merge requests that don't work (in a whole variety of ways), going really fast by opening new merge requests without fixing the old ones, local changes with no thought to the larger structure, and so on. About half the time I feel like I'm spending more time commenting on these merge requests than if I'd just done it from scratch.
I disagree. The more senior you are, the easier it is to generate something and be convinced that "ok, this is good enough" or "aight, I see these things that need to be changed instantly". Frankly, compared to others, I don't have that much experience (about 10+ish years), but it saves quite a lot of time for me.
It takes a bit time to build intuition around the workflow, but when you get going, it seems surprisingly useful. I was a skeptic before as well, btw.
I once worked at a company where the head of our department announced we were doing a re-org. He later explained that another department did so and at a board meeting the news was well received by the board ... so he was doing it too.
Sometimes leadership just feels the need to put their stamp on things so everyone knows they're there.
Elon's personality means he can't be in anyone's shadow. He needed to refound the company to remove the mental association with Dorsey and create the association that he's the founder.
(And when the answer is "someone knows how to pull his strings and push his buttons", that makes me more worried about prompt injection in organic intelligence than the same attack in artificial intelligence).
he spent half a year in court trying to get out the deal, and lost. so you probably shouldn't assume actually buying twitter was really his plan.
he bought a smaller stake first to get some influence and then offered to buy the rest when it didn't buy him the influence he hoped it would, presumably with no intention of following through. then the board called his bluff.
So this is actually a good line of inquiry; intelligent people like to assume they are invulnerable to propaganda and nudging, when this is far from the truth. It looks like he's fallen into the rightwing rabbit hole, to the extent of prioritizing particular posters such as the notorious "libsoftiktok". The persecution complex that comes along with this tends to make him view dissenting views as the work of invisible enemies.
His initial offer was motivated by a desire to be the savior. In his information bubble of American culture wars, freeing the bird from woke censorship offered him a way to roleplay the archetype of the individual hero and get his fix of narcissistic supply from his flying monkeys. To those ends, it worked. I believe he tipped his intentions to buy it quite early in a private conversation with Babylon Bee writers.
From what I understand, Elon has owned the x.com ___domain for a long time and in the past wanted to rebrand PayPal to X the same way he did Twitter. Seemingly at that point in his career there were still people around to tell him no.
He also wants to make X into an "everything" app where you'll do shopping, calls, chat with friends, send payments, etc. And if that was pulled off successfully it doesn't seem like the craziest thing to distance from the Twitter brand which has long been associated as an app primarily for news and shitposting? But he has a long way to go to get people to think of it as anything other than what it was before. Everyone on the platform is still calling it Twitter and refers to posts as tweets.
> He also wants to make X into an "everything" app where you'll do shopping, calls, chat with friends, send payments, etc. And if that was pulled off successfully it doesn't seem like the craziest thing to distance from the Twitter brand which has long been associated as an app primarily for news and shitposting? But he has a long way to go to get people to think of it as anything other than what it was before. Everyone on the platform is still calling it Twitter and refers to posts as tweets.
I feel like the way to go there would have been closer to the Meta/Facebook brand hierarchy.
X is the platform / super app and Twitter is the first app within X. Twitter accounts become X accounts but you still use them to Tweet on Twitter.
The destruction of the Twitter branding seems more out of spite than business development goals. Really the whole follow through of the Twitter purchase reeks of spite and destruction rather than building something new of value.
But I'm not a successful billionaire so I'm probably wrong.
> The destruction of the Twitter branding seems more out of spite than business development goals
Or just incompetence. Plenty of that. Assuming just because he is billionaire that he's competent at everything related to any company and the reason for failure is something else is silly.
He had no idea what the fuck he is doing and he fucked up.
Oh there is likely plenty of that too. But I think there was at least some malice in being forced to follow through on his overvalued purchase offer, in the treatment of "blue check" verified accounts, in the treatment of laid-off employees and those who remained or tried to stick it out, and in the destruction of the Twitter brand.
> Assuming just because he is billionaire that he's competent at everything related to any company and the reason for failure is something else is silly.
The only times I can think of where a re-branding exercise has worked well, have been where the brand started off in the mud and they fixed their image rather than the name.
It was before most of our times, but I don't think the Esso brand was "in the mud" when they rebranded to Exxon and that seems to have worked out ok for them.
There are similar "everything" brands that are very successful in other countries, notably WeChat which is chat, text, photo sharing, gaming and mobile payments.
If X became the first in America it would be quite powerful, but IMHO is very unlikely.
I didn't even know MS had photo sharing or mobile payments.
Google kills and relaunches products so often that I'm genuinely surprised Google Photos still exists and that my old data from when I had a Nexus 5 is still there, but I guess that's fair, I did forget them.
Google platforms like Google Pay, Google Photos, etc are very common across the developing world (India, Indonesia, Brazil, etc).
Microsoft doesn't have mobile payments but they do provide photo sharing via OneDrive.
FB/Whatsapp ofc is a massive one, and is the goto everything app globally outside of China and North America.
Most of these platforms are targeting developing countries now because any growth in market share that could be extracted in the west has been extracted.
This is why most ads made by Google, Meta, MS, etc tend to target the Indian, ASEAN, or Brazil markets now, and a lot of investment in localization is happening (eg. Meta/FB/Whatsapp partners with local telcos to install it's apps by default and integrate with telco platforms, Google has constantly massive ad campaigns in India, etc)
> Revenge for being forced to honor his word at buying Twitter, he made them regret it by killing it.
This makes no sense. The court forced him to pay the shareholders the agreed upon price. Consequently, the shareholders were paid, and now they're no longer shareholders. Why would they regret that?
Because it’s not about being a shareholder. It’s about trying hurt Dorsey’s pride. “Watch me destroy your greatest achievement!” Musk is trying to damnatio memoriae.
Dorsey was not part of the lawsuit and in fact rolled over his Twitter shares with the acquisition instead of taking the payout, so again the "Revenge for being forced to honor his word at buying Twitter" theory makes no sense.
Theoretically, there's a universe where Twitter wasn't bought out and the share price grew substantially, which will never come into place due to Elon's actions?
(I'm grasping at straws, mostly because Elon's ability to spite the shareholders is fairly limited.)
He's spoken about this in length. When it actually happened it did feel like a bit out of nowhere though, but makes sense in the engineering mindset to get it out of the way clear before fanning out to other areas, regardless of the hard-to-estimate ETA's for those features (talking payments, "mega-app"-like features, different ways of integrating Grok, possible new UI flows for YouTube/Twitch-like fronts etc)
> it's because you've run your brand name into the ground
And even then it usually doesn't work. It's _very_ unusual for full-scale rebrands of consumer-facing companies, where the original brand is utterly expunged, to stick.
he is an extreme narcissist and is very mentally unhealthy / unstable. X is his favorite letter. it's not "cool", he just fancies himself to be King Joffrey who can do anything he wants on a whim.
He bought x.com early and has probably paid stupid amounts of feed to hold on to it all these years so he's forcing it on his new pet project he totally wanted and had a plan for when signing legally binding documents. (/s)
I do believe he's just been obsessed with having x.com since he got it in the 90s and had a chance to use it and jumped at it just to put his own imprinter on the site.
The prompt was:
"Create a Python decorator that registers functions as handlers for MQTT topic patterns (including + and # wildcards). Internally, use a trie to store the topic patterns and match incoming topic strings to the correct handlers. Provide an example showing how to register multiple handlers and dispatch a message to the correct one based on an incoming topic."
https://pastebin.com/wefw7X2h
reply