faizshah

I think Elder Scrolls Oblivion was like 80 person core team and 4 years of development and New Vegas was like 70 person dev team and 2 years of development.

But starfield, which was widely criticized by fans, was over 100 people and 8 years development. With the most common criticism from former devs being about the additional management structure and difficulty of communication.

I can’t help but feel there is a lesson there for tech companies where engineering orgs for products are 200+ people when a 20-30 person startup is your top competitor in the vertical.

My favorite is 20th Century Food Court in Last Call BBS. Some of the other games remind me too much of work (I’ve bought all of them including the coincidence card games but I have only beaten this one), whereas this one reminds me of fun times I had making synths in Logic and VCV Rack for fun. Highly recommend!

I always wished they would make a management or simulation game, I think 90% of all programmers play Paradox games or Tycoon games etc. and I know their take on it would be amazing.

Not really if it takes you 15 minutes to write a 50 line function but it takes the AI 90 seconds then you already are at a 10x speedup just for this task.

This (non-yolo mode AI coding) is actually how we used to code in the old days (2023).

As someone who has used multiple vibe coded internal tools: you will care when you use these tools and encounter strange bugs and missing features.

The human touch is visible in the way your features work just like in vibe coded art and games it lacks intention.

Scale to zero is very useful.

When people say “you’re holding it wrong” it tells me they can’t even conceive of a better way of doing it.

The models produce the same slop for everybody, you don’t have a special way of doing it you lack taste and an opinion on your problem domain from lack of research and studying prior work.

If anyone else couldn’t find the working paper link (the readme and conf link didn’t work for me) it’s this one here: https://github.com/antoinezambelli/forge/blob/main/docs/forg...

It won't be an LLM that does it, the entire feature of an LLM is it produces generalizable reasonably "correct" text in response to a context.

The system that makes it have an opinion about good vs bad architecture or engineering sensibilities will be something on top of the transformer and probably something more deterministic than a prompt.

I experimented with vibe coding (not looking at the code myself) and it produced around 10k LOC even after refactors etc.

I rewrote the same program using my own brain and just using ChatGPT as google and autocomplete (my normal workflow), I produced the same thing in 1500 LOC.

The effort difference was not that significant either tbh although my hand coded approach probably benefited from designing the vibe coded one so I had already though of what I wanted to build.

The problem we had with Arcs (and Root) is that we didn’t find the strategy as deep as Dune Imperium. If you think about Dune Imperium, when you make a turn, and even between turns, you are constantly considering reveal power vs card powers vs war power vs resources, and you are making long term strategic decisions vs short term tactical ones vs deck building vs even the order of operations within a turn and round. Every time an opponent makes a move, because the game is so zero sum, you constantly have to pivot your plan if a space is taken, if a war shifts, etc.

When we played Arcs, we found it was much more about making short term tactical decisions, and there wasn’t as much room for that kind of deep long term planning, so the experience felt less competitive and less tense for us.

So we are looking for something that gives a similar player experience of constantly balancing short and long term strategy like that, but without going all the way to something with very very complex rules like Twilight Imperium.

Does that make sense?

I tried a few, the only one that really meshed with me in terms of competitiveness, strategy, constant pivoting and wide decision space is Dune Imperium Uprising. My group started with wingspan which was fun but we ended up hating how much it became just pure chance.

Dune was basically the opposite you have an element of RNG from deckbuilding, you have multi-step planning and if someone else takes your move you have to recalibrate your plan, you can pull off crazy combos and hidden plans with intrigues.

I highly recommend Dune Imperium Uprising for engineers.

Would love if anyone can recommend any game of similar depth. We tried Arcs, root, and some others but couldn’t find anything similarly competitive and deep while not being one of those way too complicated games like twilight imperium or something.

Theres also spellfix1 which is an extension you can enable to get fuzzy search.

And ON CONFLICT which can help dedupe among other things in a simple and performant way.

Personally, if I feel like you vibe coded your SaaS I’m probably not gonna pay for it. You can obviously tell when a project is vibe coded just based on the way it looks, the weird bugs you see and the poor documentation.

There’s definitely a market for good looking UI that actually works and stands out from the vibe coded junk. Artisanal corn fed UI I guess.

Dairy in the UK also tastes far better than in the US. British people often comment how hard it is to deal with the dairy in the US which tastes like water in comparison.

Rails 8.1 and ruby 3 are also very surprisingly fast, and coming back to an “omakase” framework is honestly a breath of fresh air especially now that with AI tools you can implement a lot of stuff from scratch instead of using deps.

I heard that the environment there is 996 with high turnover. So you might be paid double in comparison to a FAANG job but you work double as well. (This was about dev positions not researchers)

Anyone know if that’s true? I only heard it second hand.

One really useful usecase for Garage for me has been data engineering scripts. I can just use the S3 integration that every tool has to dump to garage and then I can more easily scale up to cloud later.

The parallel agent model is better for when you know the high level task you want to accomplish but the coding might take a long time. You can split it up in your head “we need to add this api to the api spec” “we need to add this thing to the controller layer” etc. and then you use parallel agents to edit just the specific files you’re working on.

So instead of interactively making one agent do a large task you make small agents do the coding while you focus on the design.

Not to be overly negative but I’m kinda disappointed with this and I have been a JetBrains shill for many years.

I already use this workflow myself, just multiple terminals with Claude on different directories. There’s like 100 of these “Claude with worktrees in parallel” UIs now, would have expected some of the common jetbrains value adds like some deep debugger integration or some fancy test runner view etc. The only one I see called out is Local History and I don’t see any fancy diff or find in files deep integration to diff or search between the agent work trees and I don’t see the jetbrains commit, shelf, etc. git integration that we like.

I do like the cursor-like highlight and add to context thing and the kanban board sort of view of the agent statuses, but this is nothing new. I would have expected at the least that jetbrains would provide some fancier UI that lets you select which directories or scopes should be auto approved for edit or other fancy fine grained auto-approve permissions for the agent.

In summary it looks like just another parallel Claude UI rather than a Jetbrains take on it. It also seems like it’s a separate IDE rather than built on the IntelliJ platform so they probably won’t turn it into a plugin in the future either.

I agree, for me it’s my current weekend project to try to figure out a dirt cheap and high performance self hosted cloud for hosting stuff.

So I’m still sticking with Route53 cause it’s the least annoying registrar and DNS api, for CDN I’m going with bunny and for dirt cheap object storage I’m going with b2.

Then the fun part is the actual self hosting: I’m going with Garage for my normal self hosted S3 api (b2 is for backups etc.), Scylla for DDB, Spin for super fast Wasm FaaS…

Then this weekend I got deep into trying to build my cloudwatch alternative I think I’m going with dumping logs with vector into b2 and then using quickwit for searching the logs.

Just a fun homelab challenge really.

FYI if you want an s3 + CF analogue setup, b2 is integrated with bunny and allows private buckets: https://www.backblaze.com/docs/cloud-storage-integrate-bunny...

I haven’t yet worked out the best cheap VPS/dedicated provider though, project for next weekend.

I think AI coding is another part of why this is seeing a resurgence. It’s a lot quicker to build quick and dirty scripts or debug the random issues that come up self hosting.

This isn’t true anymore we are way beyond 2014 Hadoop (what the blog post is about) at this point.

Go try doing an aggregation of 650gb of json data using normal CLI tools vs duckdb or clickhouse. These tools are pipelining and parallelizing in a way that isn’t easy to do with just GNU Parallel (trust me, I’ve tried).

I had to do something like this for a few TB of json recently. The unique thing about this workload was it was a ton of small 10-20mb files.

I found that clickhouse was the fastest, but duckdb was the simplest to work with it usually just works. DuckDB was close enough to the max performance from clickhouse.

I tried flink & pyspark but they were way slower (like 3-5x) than clickhouse and the code was kind of annoying. Dask and Ray were also way too slow, but dask’s parallelism was easy to code but it was just too slow. I also tried Datafusion and polars but clickhouse ended up being faster.

These days I would recommend starting with DuckDB or Clickhouse for most workloads just cause it’s the easiest to work with AND has good performance. Personally I switched to using DuckDB instead of polars for most things where pandas is too slow.

Can anyone recommend a good cloud for GPU instances?

Was trying to find a good one for 30B quants but there’s so many now and the pricing is all over the place.

This is well intended but I know from experience this is gonna result in you asking “how do you find and kill the process on port 8080” and getting a lecture + “Claude has ended the chat.”

I hope they implemented this in some smarter way than just a system prompt.

I've had cases where it completely misinterpreted what I was asking for, a very strange experience which I'd never had with the other frontier models (o3, Sonnet, Gemini Pro).

Yes! This exactly, with o3 you could ask your question imprecisely or word it badly/ambiguously and it would figure out what you meant, with GPT5 I have had several cases just in the last few hours where it misunderstands the question and requires refinement.

It was related to software architecture, so supposedly something it should be good at. But for some reason it interpreted me as asking from an end-user perspective instead of a developer of the service, even though it was plenty clear to any human - and other models - that I meant the latter.

For me I was using o3 in daily life like yesterday we were playing a board game so I wanted to ask GPT5 Thinking to clarify a rule, I used the ambiguous prompt with a picture of a card’s draw 1 card power and asked “Is this from the deck or both?” (From the deck or from the board). It responded by saying the card I took a picture of was from the game wingspan’s deck instead of clarifying the actual power on the card (o3 would never).

I’m not looking forward to how much time this will waste on my weekend coding projects this weekend.

o3 was also an anomaly in terms of speed vs response quality and price vs performance. It used to be one of the fastest ways to do some basic web searches you would have done to get an answer if you used o3 pro you it would take 5x longer for not much better response.

So far I haven’t been impressed with GPT5 thinking but I can’t concretely say why yet. I am thinking of comparing the same prompt side by side between o3 and GPT5 thinking.

Also just from my first few hours with GPT5 Thinking I feel that it’s not as good at short prompts as o3 e.g instead of using a big xml or json prompt I would just type the shortest possible phrase for the task e.g “best gpu for home LLM inference vs cloud api.”

The best one was the 2048 era: https://hn.algolia.com/?q=2048

So what this POS company is doing that’s kind of interesting is they have a mobile app just like Uber Eats/Doordash but it seems to be direct through the restaurant: https://apps.apple.com/us/app/local-by-toast/id1362180579

Actually when I ordered through it I paid $45 instead of $60 via DoorDash/Uber Eats but the order was delivered by Uber.

The reason why I think this version will work is because it’s the same level of convenience in a centralized mobile app so single payment method, single portal to browse restaurants etc.

(I have no association with this company I just was looking into this last week on the train)