whazor

I think wind farms on sea look really cool, especially if the turbines are huge. It gives some majestic feelings

True. Real harness engineering is when you actually measure outcome statistically. Although i would say 5 samples might be enough to conclude something

Especially with tool calling.

This could be the Opus 4.5 moment for normal humans.

My ChatGPT gives me lots of “artificial general intelligence” on a daily basis that I don’t possess.

According to my agent:

> At 6 nm: 1T doesn't fit on one wafer.

> At ~2 nm: 1T plausibly fits comfortably on one 300 mm wafer.

But then again, 300B to 500B models are to this day also very valuable

Even if its on the cloud, should be locked behind your VPN

This is a good case for custom harness/sandbox engineering.

Not my blog :)

The web_search tool and MCP discovery is essentially a global directory.

Now when I ask ChatGPT for a product, it does a web_search tool call and then gives me a link to a webshop product page. But next step would be automatically using the MCP of that webshop to embed checkout inside chat.

MCPs are very exciting. They now support IFrames (MCP Apps) as well.

Now that the models are getting better, you can trust them to book barber appointments, order groceries, manage your calendar, etc etc. It's significantly lazier to ask an agent to do something than manually doing it. Especially with voice to text.

I'm predicting that chat bots will auto-discover MCPs from trusted websites and make life much easier.

You can see the approach in his post: https://howaboua.dev/writing/how-i-gave-pi-17-tools-without-...

The result: 38% fewer startup tokens, 17 tools exposed through just three schemas, and 19 skills loaded only when needed.

Both Codex and Claude have trained their models to their harnesses.

There are some experimentations by Igor Warzocha to extract Codex shapes and put it in Pi: https://github.com/IgorWarzocha/howaboua-pi-stuff/tree/main/...

I'm expecting every model will have a fine tuned Pi extension at some point.

yes I have one and love it but I believe hardware design could be done better by Casio/Seiko

Two of my favorite (sci-)fi writers both use crazy big Excel files that almost crash their computers when loading. For research but also keeping track of story loops.

I can imagine something like ChatGPT Work with markdown/python/CSV's can do quite some magic for such writers.

There is a big difference pre and post AI, namely that humans have quite good long term context management.

But now that everyone is coding with AI, all agents need to acquire context every session. Thus the value of doing best practices is much higher and benefits are there immediately.

I wish they could build a modern Pebble time 2 watch, since its opensource

I think you can get the same experience on much more affordable VR glasses. Also installing an APK is easier than all the Xcode/Apple signing/device pairing etc.

Well I guess I have good news for you (and bad news for Dutch merchants), because Wero does offer the ability to dispute.

Yeah, you can directly print the SOTA AI model on the chip. I also believe this can be done on older process nodes. The most important factor would be speed. How fast can you go from new model to new chip?

This is why it makes sense to carbon price the private jets. On a smaller plane you can innovate much faster.

why do you need a java language server? Let them use the same tooling you would use in CI

but google sucks, its a terrible product thanks to its ads. Internet is terrible without adblockers.

Now chatgpt can add some subtle ads in their apps and they have enough users for companies to spend money on it.

As I understand it, the machine also needs lots of 'normal parts' and can be designed differently to fit the factory and the automation. Like if you need a pipe, instead of buying it from ASML, you can buy it from a local hardware vendor. And parts of the ASML also break, and those parts would be directly shipped from the vendor that makes the part. ASML manages the integration between these parts and each part can have multiple vendors that makes them according to the specs.

I should have said 'a discount' since I indeed hope they make a revised version of Time 2.

real danger of insider trading is that you can get jailed 20 years in the future for something you did today

Now with Pebble you only receive iOS notifications.

Soon in Europe you can reply to notifications via voice.

I got the pebble time 2 now for a month. Pebble is the Casio smartwatch.

The issues are a bit sad. But I will happily upgrade with the discount to the revised version of time 2.

To me, a Pebble is better than an apple watch due to battery life and hack-ability. Especially with the latest AI models it is just your imagination. And with ASK and Notification Forwarding it will be much closer to a Apple Watch in functionality.

Mikrotik also has a safe mode for testing configuration

There is also a terraform provider. Not sure if there is a safe mode here. I normally test via ssh safe mode and import the changes afterwards.

It is not. Wero has charge back support and therefor higher fees. There is quite some interest in Pay by Bank via PSD2 APIs. This is much more similar to iDeal and can work for the entire EU.

This is arguable a feature.

I made a prototype where AI automatically fills in the checkout basket for an amusement park. I found that ChatGPT tells how many adults, how many kids, what date suits you.

There are quite some security concerns, but fully banning AI from filling in query parameters with relevant user data is not the solution. This is also why I think Claude didnt give the bounty. Their solution would likely be a combination of trusted domain allow list and better security model that protects user agent.