thiago_fm

I think you may have a problem with caching, can you check how much cache % you have?

Whats your harness?

You still review code?

Maybe improve your development process, learn how to use AI better so you don't have to.

Maybe only if you are writing OS or some very specific kind of software you have to do this.

For all the other kinds of software, you likely don't need to.

This will go on until the AI CAPEX starts to drain. But there's a chance that this never happens.

All that money being spent on NVidia GPUs has to become tokens which will be bought by a business.

Many of those AI businesses are making those LLMs do something useful, which makes sense.

But right now nobody has an idea how much a single token really costs.

Once the CAPEX has to start showing some real cashflow positive numbers, it's likely that the token prices will go up, margins will compress and a lot of businesses will go bankrupt.

Of course, if this even stops. We could keep going on forever if AGI/ASI arrives, then all our previous ideas about economics which is based on human labour mostly is outdated, and we will need to figure out new economics!

This name is so good.

ZuckOff!

I wish they'd just make them illegal, it would be easier.

Like, if people want to use it on their private property, alright, legal.

But in public? I never accepted any terms and conditions for it...

Damn, this was a terrible showcase of their voice API, I really love their product. It's the fucking best, a life changer. But... what a meh showcase.

Is somebody from OpenAI hiring? I can show how I use it to learn German, among other very interesting usages.

Also show proper excitement etc... I think also a lot of real users could do better.

It feels like they aren't real users of their own products...

How about we instead make Earth be a planet good for humans and living beings?

We can slow down AI development and make it a companion for us, not a replacement.

We don't need to automate ourselves out of it. Or take the risk of extinction.

Just like with nukes, we decided to not blow up the planet. We can too change this.

GDP is a terrible measure. I bet we will see a huge drop on HDI with so many suicides and people going through hell because their job got automated or they no longer see meaning in life, feeling powerless.

Instead they should run an estimate on how many would suicide until 2030 if the 'extreme scenario' happens.

Big GDP growth and a lot of people no longer finding purpose in their lives, going bankrupt, losing their homes and family etc.

For Anthropic, you are just clearly a number. They only care about output.

They have long forgotten the human being.

Make less your life about your work. It's so hard for some people, myself included to do that. But you need to do that!

Some people don't even want to work at all or work in a successful product or whatever. We created all those silly expectations for ourselves, often they aren't even our expectations but the expectations of others.

Those expectations, our desires often make us very miserable.

For instance, 90%+ of the people in the world would dream of living in Sweden, despite your bad outlook of the job market. It's all about perspective.

Of course, if you can afford, pay a psychologist. They may help you better to deal with your emotions and help you find out what you need.

Maybe try building something... if you want, hit me up, see my profile. But maybe first try to chill down and enjoy life a bit.

Dario's messaging has been extremely confusing and HE'S RESPONSIBLE for further increasing the negative sentiment of the public towards AI.

He's clearly focusing on being on the news to provoke emotions on people, preparing for the Anthropic big blockbuster IPO.

Dario, Sam Altman and others should instead be praying every night that this will take us to the Singularity VERY SOON, because if it doesn't, the entire country will blame them for absolutely destroying their retirement and the US' economy once this AI Datacenter bubble pops.

I often wonder if there's a chance, even if minimal... that they stole the weights of the Anthropic models they run on their datacenter... or are actively destillating it.

The question is whether they can penetrate those markets before going bankrupt

My take: no

They are still pretty much competing for software development, you see only tiny verticals being tried by OpenAI and Anthropic, with Claude Design et al

This can be better done by other companies, with a fresh balance sheet, don't you think?

There'll be plenty of people at Anthropic and OpenAI opening those companies

Agents are already too good, the models improved

Nowadays you can just start with a plan.md and add everything you want there, verification etc

And you can build a loop to verify/validate it in case its a big/long-running plan

I don't have the problems you mentioned that the models aren't able to follow instructions throughoutly, they do make some mistakes, but it's up to you to set up a process that would work for your codebase to understand that the mistakes were made and need to be fixed, that can be done in a loop for example.

It's exactly this. Not only this, we might be running very soon into cheap compute soon given the current CAPEX of all hyperscalers.

DeepSeek is model that you can run anywhere, meaning that any datacenter/company could run it at a price that would give them a margin taking in account running costs of the datacenter + depreciation.

I'm very sure we're overbuilding capacity, and we'll find out around 2027.

Those open weights will become extremely attractive because all that datacenter extra capacity laying idle will be possible to be used by builders.

It may be the case that in 2027-2028 will be cheaper to start a new AI model company from scratch to compete with Anthropic/OpenAI as you'll be able to:

- the new company won't be tied to so much capital, debt and running costs

- get much cheaper datacenters to train than it is Today, as the big labs will be stuck with bad infra contracts

- have great engineers/researchers willing to jump ship because if Anthropic/OpenAI valuation falls, that will destroy their equity

- wide available chinese models to distill from, papers and ideas on how to acquire datasets

Anthropic/OpenAI are playing an extremely risky game, they need to greatly keep doubling their bets and delivering model big performance increments.

If they don't give us ASI/AGI in 2 years or find huge new markets, there's no way the economics will make sense.

The execs forgot why AI models even exist (like huge usage for the cyber models, or new types of models)...

AI models' usage is mostly tied to build Software Today, that's where the value really lies now.

If there isn't enough Software to be built as people claim to the rate they expect, eventually the supply of AI models will outstrip the demand.

You can't sell pick & shovels infinitely, AI models themselves are also in that category.

Eventually you need enough holes to be dug, and gold to be struck, and nobody knows what's that demand. (even though they built all that supply...)

If you add all those .md files to your AGENTS.md, your harness will likely load all those files into context. This will create context pollution for you.

Context, similar to your computer's memory, is limited.

Additionally, the more things the AI has in context, the worse it performs.

Skills helps you to quickly add more context for doing things, without polluting your context window.

I think you need to learn about AI, there isn't anything hype there. It's just an easier way of giving more context to the AI.

This can be helpful, for example, if you have an internal API that does XYZ, and you want the AI to know that API, you can save that API, its endpoints and authenticaiton methods to a skill, and invoke it when you want your harness (CC, Codex) to know about the existance of that API, to make requests, make tests etc.

Customer Service.

You could scope it down to a demo. Share with people and see the results

If good, continue with it, if bad, give up.

Demos are so good for that purpose, if you show the game loop and it sucks, you'll need plenty of work to change this situation around (if even), so try it.

Wish they'd be a bit more open source about this, bet people would contribute more.

Also about privacy-first apps.

As if all progress done in open models is because of distilling...

People have no idea and everybody pretends to be an expert and ignore how good China is on AI research

Of course, for example, a Civil Engineer learns integrals, so it can learn advanced engineering topics.

It may not often use them often as it uses software, or it may not remember all integrals rule tables, but it understands the concept, it's usefulness, and in case he needs it, he can figure out a way to work it out.

This is the same for Software Engineering and code. You don't need to have written the HTTP protocol library in order to use it.

I haven't written a single line of code since more than an year, but I've made the AI write multiple thousands of lines of code since then.

It's just a calculator. You still need to know how to use the calculator and for what purpose do you use it.

What changed is that you no longer will need to learn all ins and outs of coding.

Just wait until they release their coding model. Once they do an Opus-level coding model, the sandcastle of the AI economy in the US will fall

Great. Wish we've had more people in the community thinking about solutions like this.

Why should only kids be protected from addiction?

I have a hard time understanding this.

We have plenty of adults with terrible social media addiction that is destroying their lives, and nothing being done about it.

This is a clear display that we need free trade, sensible economic polices and a common ground of what humans need to thrive. "Sovereignty" is overrated.

For example, for the US to have a chance in the EU, it would first need to fix its YOLO fiscal policy of sustained 5.5% debt/gdp deficits.

We shall see in a few years as US's debt balloons and the average American becomes pseudo-slaves from a few overlords... to see if the EU is really bad as some Americans believe it to be.

Yes and no.

I love working with AI and agents.

But I don't know how long this will go, so I'm investing on getting fit again for random jobs like moving atoms.

Or realizing my life's dream: become a lifeguard in public swimming pool or learn the trades to maintain swimming pools. There's a shortage of them where I live and it's enough money to pay for food etc.

I love swimming and watch people happy enjoying the water, it's such good vibes I'd do this forever.

Our life is mostly suffering and grinding through... but people on swimming pools are always smiling, having a good time & having the basic pleasures of enjoying the water and the moment.

It isn't surprising at all, Microsoft is doing a PE firm playbook with what they buy. You don't need to look much far, let's think about its biggest acquisition to date, Blizzard.

Blizzcon canceled. All of its IP barely got any love.

See what players think about the latest World of Warcraft patch. It's absolutely shit and broken. People say they fired the entire QA department since a few years back and since then the quality has just gone down.

They buy those businesses because they have nothing to do with that free cash flow, and for accounting reasons it makes sense to have them.

They didn't buy those businesses to develop it further and make it worth more.

Github will just become ever more irrelevant.

The key issue is that the US governments let those huge monopolies exist, and then use their money to buy other businesses and enshiftify them.

Unless that changes in the US, this will continue happening.

Beautiful answer. Priceless!

LLMs for language feels like it's definitely the way to go. I feel like that by just improving it further can definitely reach perfection, if not very close.

My concern is mostly all adjacent fields, like systems thinking, spatial reasoning, "real" human-like reasoning etc or as you put it, "AGI".

Doesn't seen this will take us there at all. I don't feel like we're closer to AGI than we were on the earliest versions of ChatGPT.

Thanks for your perspective from somebody working on the field. I still wonder though: what would the results be if we'd just use a richer dataset + more parameters? Would it be really that different results? (except costs, as MoE def helps with that)

MoE: I assume some people just specialize in working with routing as with that, as by reducing the amount of params and just using a subset, you end up making it less costly. So, AI researchers are only working on optimizations on getting this better?

Same question on Reasoning, so AI researchers are working mostly on optimizations on top of it, like CoT and so on, like mini-optimizations.

So basically, they work on those micro-optimizations, put them together and see a % improvement in a benchmark?

I'm sure this is probably awesome for languages, which if I'm not mistaken, it was the use-case initially used on "All you need is attention" and the entire LLM revolution.

But this seems to be a very clear path to be "taking the car to the carwash by foot" for a long time, isn't it?

It feels like we'll keep "taking the car to the carwash by foot" until somebody optimizes for that prompt, or some pre-training done, and then there'll be another prompt that will show that the AI has real trouble with very basic real-world reasoning and imagination.

Isn't it the case, or do you see any kind of research that could take us from that plateau full of micro-optimizations that get us a few cm higher to the peak?

I've had a similar problem with other hardware, my solution was to lower my expectations of "having a shop" to fix it.

No shop will want to have the expectations to successfully fix a complex GPU, unless it's their main business, as the solution can potentially not be 'reballing the memory modules' and they might just break your GPU in the process, they don't wanna have to pay you back in case they do, otherwise the risk they are undertaking isn't worth the chance of they fixing it and getting your X euros.

Go to hardware communities and ask them if they'd like to have a go at it, find somebody reputable and go with them.

Otherwise based on what you said, you literally will have to trash it, or find a way to constantly cool it down in a very reliable way (Good luck! I tried the same with a CPU with an overheating issue and failed miserably. Hopefully you'll manage it)

Sorry for the delay, I do see people changing their behaviors.

A lot of them have burned out and were/are unemployed. Great engineers and all.

The ones that are employed, are grinding like there's no Tomorrow.

This changed a lot from 10 years ago, when it was much more relaxing to work with software.

Japan also obviously also monitors this.

https://nerv.app/en/

This kind of data is actually shared by governments with each other as well.

Science has no borders, much less disasters.

If nicbou the builder is having a hard time, that's all the economic downturn data point I need.

I share a similar sentiment about Germany. I mean, we do have a recession for a couple of years already.

As a Software Developer, I've experienced layoffs of International companies just nuking their German team, for both cost and law risks (from people trying to create a worker council and the like).

I'm still employed because of my YOE, my skills & a wide network of people that have seen the quality of my work, but I see even previous CTOs and great engineers without a job.

Maybe in the years ahead, I might need to work doing something else.

I've always wanted to go to trade school and run my own business anyways, just didn't due to software engineering being so fun, interesting, challenging and ofc, well-paid.

I've been practicing my German a lot (C1+), so in worst case scenario I can do other work, maybe become an electrician or something that involves moving atoms, rather than bits.