lexandstuff

Yes, we're seeing issues with Dynamo, and potentially other AWS services.

Appears to have happened within the last 10-15 minutes.

No, this violence is from a deranged person who shouldn't have had access to a firearm in the first place. No culture wars please.

I find it useful. A nice little tool in the toolkit: saves a bunch of typing, helps to over come inertia, helps me find things in unfamiliar parts of the codebase, amongst other things.

But for it to be useful, you have to already know what you're doing. You need to tell it where to look. Review what it does carefully. Also, sometimes I find particular hairy bits of code need to be written completely by hand, so I can fully internalise the problem. Only once I've internalised hard parts of codebase can I effectively guide CC. Plus there's so many other things in my day-to-day where next token predictors are just not useful.

In short, its useful but no one's losing a job because it exists. Also, the idea of having non-experts manage software systems at any moderate and above level of complexity is still laughable.

And all of those companies except for Google are entirely dependant on NVIDIA who are the real winners here.

The trend of offshoring came and went nearly two decades ago.

Time zone differences, language barriers and cultural differences proved insurmountable.

Hybrid remote seems to work quite well, on the other hand.

The point of wearing a mask is to protect other people from your respiratory droplets. Please wear a mask when you're sick.

Searching the course Slack channels looks like it doesn't run on Linux, nor a VM (though I'm sure with enough fiddling the latter can be done)

I had a look at another student's diploma certification, and it says:

> "... having registered with the University of London and passed the approved examinations conducted by Goldsmiths College, has this day been admitted by the University of London to the Degree of BACHELOR OF SCIENCE".

So I don't think it would be unreasonable to mention Goldsmith's College alongside UoL as your education history.

Yep, all exams are done remotely, using Inspera proctoring software.

The "classes" are just videos on Coursera - totally optional. You have mandatory midterm assignments, and either a final assignment or an exam, depending on the class.

Hey friend,

Happy to answer. Feel free to get in touch on one of the socials in my profile to discuss further.

Q1. I'm from Australia, and the target destination is USA. Although I have no firm plans, I want to have the option in future. Australian's have access to a particular visa called E-3, which I have failed to obtain in the past due to lack of degree.

Q2. Yes. Upon completing this, I will have a BSc in Computer Science from the University of London. It's a bachelor's degree as good as anywhere. The exams are proctored, and all the exams/assignments are graded via Goldsmiths University. It's 100% legit.

Q3. You pay per module, so you don't need to pay up front. You just pick how many modules you want to do that semester and pay for them. The only requirement is that you must complete enough modules per semester to finish within 6 years. Since the education is directly related to my work, it is also tax-deductible in my country.

I'm a CS professional of 15 years, getting a CS degree for visa purposes. I'm doing it via the University of London's World Class on Coursera [1], and it costs around ~20-25K USD for the whole thing. If all continues to go well, I'll be done in 3.5 years, all while working full time.

So that's another option to consider: do the CS degree part-time while working on breaking into the industry. You can adjust your workload each semester depending on circumstances, so it's flexible for someone job hunting.

[1] https://www.coursera.org/degrees/bachelor-of-science-compute...

I don't fault you for maintaining a healthy scepticism, but per the President of the IMO: "It is very exciting to see progress in the mathematical capabilities of AI models, but we would like to be clear that the IMO cannot validate the methods, including the amount of compute used or whether there was any human involvement, or whether the results can be reproduced. What we can say is that correct mathematical proofs, whether produced by the brightest students or AI models, are valid." [1]

The proofs are correct, and it's very unlikely that IMO problems were leaked ahead of time. So the options for cheating in this circumstance are that a) IMO are colluding with a few researchers at OpenAI for some reason, or b) @alexwei_ solved the problems himself - both seem pretty unlikely to me.

[1] https://imo2025.au/wp-content/uploads/2025/07/IMO-2025_Closi...

Personally, I think it's a perfect word for what it is: carelessly created content that no one wants.

It used to be that if I saw a typo or grammatical error in someone's writing, I'd switch off, thinking the author really didn't care that much about the text they're writing to proofread it. Now, the complete opposite. Leaving in typos and such shows a clear signal that the author cared enough about what they're writing not to out source to AI.

Related to that, I saw a local band posting marketing material online, that was this kind of amateurish typography with a collage of photos decorated with coloured markers. 2 years ago I'd be laughing at what a terrible job it was, today, it's a breath of fresh human air from all the slop we're subjected to all over the internet. It caught my attention, so much that I'm going to see the band this weekend.

Out of interest, has anyone ever integrated with Grok? I've done so many LLM integrations in the last few years, but never heard of anyone choosing Grok. I feel like they are going to need an unmistakably capable model before anyone would want to risk it - they don't behave like a serious company.

Sometimes you just have to deal with the world as it is, not how you think it should be.

Yes, but you're missing their ability to interpolate across that dataset at retrieval time, which is what makes them extremely useful. Also, people are willing to invest a lot of money to keep building those datasets, until nearly everything of economic value is in there.

And there's a huge increase (apparently around 36%) in road fatalities during bad weather.

People have much higher safety standards for self-driving cars than they do for human drivers. Just look at how one fatality led to the total abandonment of both the Cruise and the Uber self-driving program.

It surely means these things will never operate safely in rain, snow or even bad fog. That seems like a pretty big issue for a taxi operation.

Maybe, but there's also some evidence that AI coding tools aren't making anyone more productive. One study from last year found that there was no increase in developer velocity but a dramatic increase in bugs.[1] Granted, the technology has advanced since this study, but many of the fundamental issues of LLM unreliability remain. Additionally, a recent study has highlighted the significant cognitive costs associated with offloading problem-solving onto LLMs, revealing that individuals who do so develop significantly weaker neural connectivity than those who don't [2].

It's very possible that AI is literally making us less productive and dumber. Yet they are being pushed by subscription-peddling companies as if it is impossible to operate without them. I'm glad some people are calling it out.

[1] https://devops.com/study-finds-no-devops-productivity-gains-...

[2] https://arxiv.org/abs/2506.08872

Great article. The other thing that you miss out on when you don't write the code yourself is that sense of your subconscious working for you. Writing code has a side benefit of developing a really strong mental model of a problem, that kinda gets embedded in your neurons and pays dividends down the track, when doing stuff like troubleshooting or deciding on how to integrate a new feature. You even find yourself solving problems in your sleep.

I haven't observed any software developers operating at even a slight multiplier from the pre-LLM days at the organisations I've worked at. I think people are getting addicted to not having to expend brain energy to solve problems, and they're mistaking that for productivity.

Well, they just had a $3B exit, so not that grim, all things considered.

This is a court order. They saw them in court, and this was the result: https://cdn.arstechnica.net/wp-content/uploads/2025/06/NYT-v...

Even role prompting is totally useless imo. Maybe it was a thing with GPT3, but most of the LLMs already know they're "expert programmers". I think a lot of people are just deluding themselves with "prompt engineering".

Be clear with your requirements. Add examples, if necessary. Check the outputs (or reasoning trace if using a reasoning model). If they aren't what you want, adjust and iterate. If you still haven't got what you want after a few attempts, abandon AI and use the reasoning model in your head.

Not to mention, if it's an ML workload, you'll also have to factor in downloading the weights and loading them into memory, which can double that time or more.

Yes, that's basically correct. Except be warned that the cold start times can be huge (30-60 seconds). So scaling to 0 doesn't really work in practice, unless your users are happy to wait from time to time. Also, you also have to pay a small monthly fee for container storage (and a few other charges iirc).

Microplastics have recently become understood, but humans have lived with a lot of plastic for many, many decades now.

Fair enough. I guess I was mainly thinking of how rarely I need to utilise the old prompt engineering techniques. Stuff like: "You are an expert software developer...", "you must do this or people will die." etc

I just tell the AI what I want, with sufficient context. Then, I check the reasoning trace to check it understood what I wanted. You need to be clear in your prompts, sure, but I don't really see it as "prompt engineering" any more.

Isn't SWE-bench based on public Github issues? Wouldn't the increase in performance also be explained by continuing to train on newer scraped Github data, aka training on the test set?

The pressure for AI companies to release a new SOTA model is real, as the technology rapidly become commoditised. I think people have good reason to be skeptical of these benchmark results.

Prompt engineering isn't really that important anymore imo. If you're using a reasoning model, you can see if it understood your request by reading the reasoning trace.