iagooar

My kids less than 10yo both, each have already created their first video game through prompting. It is actually pretty amazing to watch them explore their creativity - and how they interact with AI is fun too. The younger one always starts with "Dear AI, I want you to..."

If the progression of AI continues, we might actually have LLMs re-create games from scratch just through watching them. They can already createe pretty accurate games from memory, imagine what happens if they get 10x as intelligent and capable.

It's exciting and kind of scary.

A few years ago I hurt one of my shoulders pretty badly, so I could barely move and had to sleep in a sitting / reclined position for almost a week. No keyboards, no computers.

Boy oh boy did that LotR premium extended something-something edition on DVD turn out to be a godsend with the countless hours of behind the scenes footage and interviews, and director's commentary. I watched it all.

I wonder how the industry has not yet learned to use ASD-STE100 Simple Technical English for documentation. It makes it so much more polished and so much less AI-sloppy.

Not my experience at all.

As much as I love my CLI tools, I just cannot go back to not having native built-in browser integration, links, etc which the (Mac) desktop apps like ChatGPT or Claude offer.

I do use Hermes sometimes, but it tends to keep growing skills and overhead over time, to the point it is becoming utterly slow and sluggish. The Desktop experience itself seems vibecoded without taste, in contrast to the core of Hermes, which is solid.

I wonder if there is an agent harness that has the strengths of Pi, but in a native Mac desktop app packaging?

I am also seeing slower speeds, roughly the same ballpark, sometimes even lower - 10-11 tok/s on M5 Max. If there is that ONE version (GGUF or MLX) that runs roughly as fast as 3.6 used to run, please let me know.

What would be incredible is the 3.8 35B MoE version too, I can run 3.6 with 60 tok/s which is a really, really nice speed.

My observation is that people who are smart (or think they are) tend to express themselves via trying to be overly correct - on everything. And that gets exhausting.

People who try to be too precise and accurate, pay it with their happiness and peace. There is no better therapy than bullshitting with friends, just rolling with it instead of trying to make it all 100% perfectly accurate.

10 years. I still have the game in my Steam library, waiting for me to try it.

I guess I will never try it, once you hit your 40s, games lose their appeal anyway.

I am working on a framework that allows creating very complex SaaS products with AI, that actually work. Very complex = +100k LOC of one-shotted AI-generated code. Codename o2p - outcome-to-proof.

It is a research-first method for designing or reviewing SaaS products. Its central question is: What customer outcome matters, and what evidence proves that the product delivers it?

Idea or existing product → Research real customers and current work → Identify their Jobs to Be Done → Design the minimum safe workflow, define the interface and system behavior (with a sophisticated scoring system) → Validate the workflow → Implement vertical slices → Observe proof of customer progress

I love DeepSeek V4 Flash since the pre-0731, now even more. It is the first model that is truly too cheap to meter.

But I find it having a pretty significant problem with tool calling - no idea why, but tool calling with it is SLOW. As long as the model is reasoning, all good. But give it a bunch of tools and it becomes extremely slow.

Am I the only one experiencing this?

Not sure I agree. It might sound like theater if you expect it to solve your security issues. But it is extremely good at finding undocumented or broken processes - these processes tend to hide the biggest offenders.

So far every single test I've done has worked correctly, I put a lot of work and time into preparation and planning, each workflow was designed, researched and optimized before implementation. The specification was already a massive document, so it is not a one-shot by no means.

I might check it and if it turns out to just be extremely bloated, I will re-write it from scratch. But before I do that, I might as well give it a chance.

No, LLMs have not plateaued, and each of the latest releases has been a proof of that.

Fable 5 is a model you instantly FEEL how smart and superior it is. GPT 5.6 Sol is a HUGE incremental improvement in multiple directions and dimensions.

DeepSeek v4 Flash 0731 is a huge improvement over the preview version, using exactly the same architecture.

Kimi K3 gets open weight models very, very close to the frontier.

No my friend, we are not done yet.

Having invested in a machine with 128GB of RAM, I would love seeing something a bit larger than 27B / 35B, possibly a 54B dense model or 70B MoE would be much closer to the Qwen 3.8 Max experience.

What is your story, how did you get into robotics? Was it your first choice?

I need to use this thread to mention the Polish payment system Blik. It is one of the most convenient payment systems out there - you generate an ephemeral 6-digit code in your banking app, copy it to the vendor, confirm, done. Works P2P with phone numbers, too.

Interestingly, it would be the perfect system for payments using AI agents - the agent tries to pay, asks user to provide 6-digit code, submits payment request. User confirms via banking app. Done.

Scanning mailbox, reading and classifying emails. Scanning a knowledge base, reading and improving individual articles, reading support interactions and creating summaries, checking and researching new leads that signed up. So much that is possible.

On paper the M4 should be roughly 1/3 of the M5, in practice it is only 1/2. With the right, optimized model like qwen3.6 35B MoE MLX you can get over 40 tok / sec on it. I run dozens of background jobs that are not time-critical on it.

I am not going to flag you, I am much OK with having good arguments.

I just purchased a Mac Mini M4 Pro 64GB for $3k - 2nd hand of course.

I am not a hater of Nvidia and I am planning on building a workstation based on RTX cards. You clearly do not seem to understand how convenient the MacMini actually IS - the form factor, how quiet it is, how durable it is, how well it integrates with other Macs, how well it works as a bridge to a personal agent like Hermes (integration with iMessage, Calendar, Reminders, iCloud, etc).

I am pretty sure I know a thing or two about computing, I have been in the trenches for many, many years and I have had machines of all kinds, shapes and colors. It just so happens that Macs are very capable, very convenient machines that happen to work great in the era of LLMs, too.

But you do you.

Buy a refurished or 2nd hand one.

I disagree LAN connection is the bottleneck. I do even work with it remotely via Tailscale on shaky hotel WIFI and it works fine (or as fine as any other API-based model).

Just buy a Mac Mini really is good advice if you want to get into real, always-on convenient agentic work.

Soon it is going to be good even for coding using local LLMs. Until then, just run API models on it for coding, local LLMs for "knowledge" work or daily driver agent like Hermes.

I thought they might ship an M5 Max version, but you are probably right.

qwen3.6 27B MLX 8bit -> 15 tok / sec. A bit slow but it is a delightful model to use, and smart too.

qwen3.6 35B A3B MLX 8bit -> 85-90 tok / sec! It is impressively fast and roughly 90% as good as 27B (in my opinion).

My problem is I won't accept anything lower than the 96GB the RTX Pro 6000 Blackwell has. My dream is a workstation with 2x Pro 6000 to run DeepSeek v4 Flash comfortably, possibly qwen 3.6 / ornith on turbo speed.

But man, I have never purchased a computer which is more expensive than a decent family car.

Ballpark 25-30 tok / sec on the Mac Mini Pro M4 + qwen3.6 35B. The generation itself is good, prefill is known to be slow on any Apple M-chip architecture. It is really decent.

M5 Max. But I also have a MacMini M4 Pro 64GB. Qwen3.6 runs on the M4 just fine - sure the M5 is at least 2x the speed. If Apple launches a MacMini with an M5, I will be the 1st one to get it.

Get a 2nd hand one. I was lucky enough to get a new one first, last week I get a 2nd hand one in order to run one of my Hermes minions at work.

I love my MacBook Pro M5 128GB RAM and I love qwen3.6.

BUT DO NOT buy this MacBook if you plan on doing serious coding using local LLMs with it. The reason is simple: your fingers will burn and your head will explode from the noise.

Running any kind of sophisticated job on the very laptop you are using is just not viable. Sure you can use it in clamshell mode, but forget touching it while working with AI coding or agents.

If you want to run Qwen3.6 27B / 35B at its best, get a MacMini M4 with 64GB of RAM and put it in the basement - or at least a few meters from your desk. Connect to it over LAN or Tailscale. The MacMini will also cost you almost 1/3 of the MacBook Pro.

Thank me later.