I've been really impressed with how good Cursor is at coding. I threw it a standard backend api endpoint and database task yesterday and it generated 4 hours of code in 2 minutes. It was set to Auto which I think uses some Claude model.
When I visited a local prison through the https://www.douglassproject.org/ I had this exact thought: why not allow remote work? I'm glad it is being done somewhere! I hope it becomes more commonplace.
I'm a technical co-founder rapidly building a software product. I've been coding since 2006. We have every incentive to have AI just build our product. But it can't. I keep trying to get it to...but it can't. Oh, it tries, but the code it writes is often overly complex and overly-verbose. I started out being amazed at the way it could solve problems, but that's because I gave it small, bounded, well-defined problems. But as expectations with agentic coding rose, I gave it more abstract problems and it quickly hit the ceiling. As was said, the engineering task is identifying the problem and decomposing it. I'd love to hear from someone who's used agentic coding with more success. So far I've tried Co-pilot, Windsurf, and Alex sidebar for Xcode projects. The most success I have is via a direct question with details to Gemini in the browser, usually a variant of "write a function to do X"
From an anthropological standpoint, consider that most every human culture in across time and space has people who play the role of the prophet. They tell us of the future. We are no different.
It is if you are hosting; but if you are going to the party...hey, it's free food! I think a systematic analysis would show that it would be cheaper for all of us on the whole to share food at parties since it is cheaper to buy in bulk.
An iOS app which connects parents with their children's screen time via screenshots and AI. Makes your kid's screen as visible as the living room TV. When screenshots are off, you choose what to allow; everything else is blocked. When screenshots are on, you choose what to block; everything else is allowed.
I have found AI generated code to be overly verbose and complex. It usually generates 100 lines and I take a few of them and adapt them to what I want. The best cases I've found for using it are asking specific technical questions, helping me learn a new code language, having it generate ideas on how to solve a problem for brainstorming. It also does well with bounded algorithmic problems that are well specified i.e. write a function that takes inputs and produces outputs according to xyz. I've found it's usually sorely lacking in domain knowledge (i.e. it is not an expert on the iOS SDK APIs, not an expert in my industry, etc.)
"Reading other people’s code is part of the job. If you can’t metabolize the boring, repetitive code an LLM generates: skills issue! How are you handling the chaos human developers turn out on a deadline?" Good point! Treat AI generated code as if somebody else had written it. It will need the same review, testing and refactoring as that.
It makes me wonder if we really so different than those who carved metal images to worship thousands of years ago. I made a 30s video exploring this idea here: https://www.youtube.com/watch?v=B7RoeHHqnAM
Is there a performance benefit for inference speed on M-series MacBooks, or is the primary task here simply to get inference working on other platforms (like iOS)? If there is a performance benefit, it would be great to see tokens/s of this vs. Ollama.
Yes! I've often said "software engineers should be doomed to use what they create. Or at least watch others try to use it." One example is our local Costco parking garage. They replaced the old push-button ticketing kiosk (which had nothing wrong with it) with one that had a touchscreen. Many times the line is backed up and one day I saw why. The guy was pushing the touchscreen button as if it were physical, and it wasn't registering the tap. He was using multiple fingers and mashing instead of using one finger and doing a clean tap inside the digital button.
It's not just books; most websites technically don't allow scraping content, but most of the content on which these models trained was scraped from the web. It's legality is still an open question.
It is shown running on 2 or 4 raspberry pis; the point is that you can add more (ordinary, non GPU) hardware for faster inference. It's a distributed system. The sky is the limit.
As others have said, a high powered Mac could be used for the same purpose at a comparable price and lower power usage. Which makes me wonder: why doesn't Apple get into the enterprise AI chip game and compete with Nvidia? They could design their own ASIC for it with all their hardware & manufacturing knowledge. Maybe they already are.