moezd

VPN is a legit technology with real applications, not a toy you spin when lawmakers hurt you. And it's for Anne Frank's diaries out of all disputes?

EU lawmakers follow tech trends by a lag of at least a decade and lay waste as they move along. Imagine what they will require once they understand how tech people no longer use chatbots and how much we can control AI agents.

That doesn't sound good, but out of curiosity: How do you exactly hide debt? Circular economy is easy to understand, but what else is cooking?

I doubt it will ever become a $1T company ever again.

Imagine building your solution using a 3rd party library, it still builds during your PR reviews and when you merge, it no longer builds. How would you feel?

Yes, and you can also write two different sentences using two different pens in your hands.

So, how do their overall financials look these days?

With tools like GitHub Actions and some added constraints, it's not always possible. You literally need a commit to trigger the CI workflow and it starts to trash your branch. Besides, aren't we all familiar with git commit -m "typo"?

guess they already ran out of fresh context huh Coming up next: You harness takes daily backups of your entire disk and restores.

Hard agree. I used to have a tuned setup where I could force it to do research properly, summarize in chunks that it would remember and form the synthesized response that way. Nowadays it's just like "oh I forgot about using that tool, sorry", "yeah I know we agreed on that and I didn't do it anyway", "That knowledge is beyond my training date, I suspect foul play" - even when you instruct it to fetch latest info all the time, or "you already told me X. This cancels your reasoning about A, B, C, so D is the only logical choice", even when those clearly still have merit, oh and never ending "your previous discussion X is relevant here, in combination to Y, but not so much as Z since there's a OSS implementation of it and another one blablabla..." Like, who remembers these all at the same time in their heads?

It's like with each release they force you to reconsider your pipelines altogether, and without announcing changes properly, you feel like a junior JS developer fighting dependencies once again.

I couldn't find architecture.yaml in the repo.

This. And don't forget the occasional refinement meetings during which you think you're making a group decision but your manager and the other senior (both German, speaking German, having spoken in private without your awareness) already decided on many things, even though they acknowledge that you're the more experienced person in the room. So you're treated as a rubber stamp and three months in when you have an objection because clearly you're walking into a sunken cost, will be shown ADRs (in Denglish) basically where everything's already decided in their favour.

Agents struggle with this: DRY means they create helper functions and classes for a testing scenario and maybe it's called two times. When you ask for a refactor of your code and tests to follow, these helper functions are also ignored and then, dead code starts piling up. Specific cleanup sessions always seem to leave residues behind and doing Ralph loops in the first place seems to help with this, but I'm just not satisfied with the overall performance. Any ideas?

Oof. That's a bright red flag: Regulated industry and own laptop? Please tell me at least if they asked you to install a new harddrive and company VPN. If not, the company is probably compliance theater and has no business being in the business. God have mercy on their customers.

How to handle it? If you can afford it, leave and don't come back. If not, bring it up first thing you see your manager, latest on your first 1-1. This can't go on. If he brushes off, check if he intends to work you hard like 996. That usually means they are looking to use your work and find a reason to fire you asap afterwards.

Just out of curiosity: Did you oversell yourself during the interviews?

correlation vs causation https://www.tylervigen.com/spurious-correlations

Chinese whispers, simulacra... I don't have the energy to argue after being name called, but you get the point. Yes LLMs are useful in building automatic telling machines, but ask it to do anything more substantial and all you are doing is burning tokens at the altar of Anthropic and hope. That just doesn't fly in regulated industries.

Intelligent humans are capable of following diverse and intricate analogies and draw lessons from seemingly unrelated events. Try asking an LLM to summarize an article and use an imprecise way to state your view. Ask it to push back. You will be drawn into so many pedantic arguments that burn through your tokens within a few messages, you'd wonder if there's someone deliberately taking over the keyboard on their side and spending your token limit. This would never happen with an intelligent human being unless they have nothing better to do and want to troll. This is a speech pattern that LLMs are trained on, it's not a show of intelligence. This also applies to LLMs claiming consciousness: The internet is full of people writing about sentience, talking to "superior aliens" in blog posts, forum threads etc. It's the speech pattern that's copied, not actual thoughts and feelings because LLMs perceive, suffer, have aims or dreams...

Let Rust vs C battle commence with renewed haste!

LLMs are still next token predictors, just because you can give it more vague instructions and it still finds the right steps to follow, it doesn't mean it's intelligent. It means you're speaking the same language as the harness they trained your model on.

And that has a limit. If you are stuck at PoC level or simple apps, you have no idea how limited the current models still are. There you really need to break tasks down, not just trust a token predictor to list steps that sound good. There has to be a human in the loop somewhere, because by the time you start skipping permissions, best case you get the jackpot, more likely is you get a suboptimal solution and token waste and what's genuinely still terrifying when the model ignores instructions and does some stupid nonsense, ruining your day. It really is as sharp as a CNC machine. It's not not useful, but could be dangerous, so maybe don't try to carve wood with a monster machine, or park your Ferrari in that crammed neighbourhood if you don't know how to parallel park.

Memory used to be the bottleneck because it was the slowest after SSD and NVMe arrived. Now they became the bottleneck again but for a different reason. One has to wonder.

Sometimes they are just too addicted to gossip by the water cooler. For all the information that flies in and out of different teams, departments and organisations, water cooler is probably the best place to learn enough to position yourself in the company. A narcissistic boss would be obsessed with that, naturally, and consider others who don't as outliers.

Empowering speech for all beginners in all fields for sure. If you're struggling today because LLMs seem to eat your lunch in one way or another, it's a good feeling to remember this.

And yet, it's also a sign of how far we're going down the rabbit hole of trusting next token predictors to do everything for us. No amount of harness, allowing it to complete tasks by matching the templates it memorized, should convince anyone that LLMs have novel ideas, because it never will. Stop publishing your own framework's code on the Internet for six months and it will diverge, always producing legacy code. Stop writing your latest spicy analysis on international diplomacy and it will continue to sound like the hopeless optimists that we all once were last year.

LLMs are golden mean generators. They will continue to rehash what's genuinely useful out there while being far from inspirational. It will get your job done, probably, but won't shock and awe people, let alone experts.

That's maybe wishful thinking from my part, but more towards like other engineering fields: Project engineers design it from scratch, everyone must speak architecture, customer and compliance at once, and we will have standards and "codes" drawn up by the end of this decade.

Seemingly similar mechanism here: https://en.wikipedia.org/wiki/2023_Turkey%E2%80%93Syria_eart...

Devastating stuff. Hopefully they get help soon. First 72 hours are critical.

SpaceX + Twitter + XAi + Cursor is the perfect combo to collect government contracts. You want exposure in social media? Use Twitter. Civil servants want AI? Use Grok. Developers want AI? Use Cursor. All space heading business? Contact SpaceX. All under the polished (!), totally unblemished (!) PR image of Musk.

Someone please wake me up if this is a fever dream.

If it's code that you can tolerate being somewhat messy and suboptimal, you can run agents e2e. If it's critical piece of code that has become part of your identity, better do the PR work and scrutinize it well. LLMs are still next token predictors, no matter how much harness, hooks, skills and tools is attached to them. LLMs will only know that these are callable, interpretating the state and mitigation are still best effort.

Everything is designed around collecting your data somehow so that they know what to sell you next. You've no longer the customer, you are an agent that gets to use that interface.

Intelligence explosion won't happen if we keep on investing big cloud LLMs and people who are in the business of being in busines (cough Musk cough). It will arrive when AI penetrates so much deep into our lives that we consider quality local models like having electricity at home, thus we rethink what it means to truly integrate AI into everything.

Do you have access to Anthropic model weights to run them locally?

South Sea Bubble part 2. I guess that's one way to break the gridlock and make history, as much as I hate to admit it.

Not yet. Without pure Apple game or decent GPUs, even with a lot of RAM and threads, all you get is about 30-50 tokens/second, and that's thinking turned off. Without these optimizations your model will have a field day with your MCPs, skills and agent descriptions and you will watch the paint dry before seeing the first output token. Local model serving means you have to fight for every token in your context window, which is quite opposite of what Claude/GPT/Copilot are pushing the industry towards.