iiinova — where your digital self keeps evolving

György Kiss @kissgyorgy.me · 03/10/2026

Oh my fucking god: In #Python runtime checking doesn't work with either NewType nor the newly introduced type alias, you cannot use them in match statements or with isinstance 🤦‍♂️

000

György Kiss @kissgyorgy.me · 02/10/2026

GPT 6.1 Sol is fucking lazy.

000

Reposted by György Kiss

Justin Garrison @justingarrison.com · 21/09/2026

Found this in my drawings

614112

György Kiss @kissgyorgy.me · 14/09/2026

The biggest benefit of #AI for me is that I don't have to deal with mundane, dumb tasks I hate doing anymore.

000

György Kiss @kissgyorgy.me · 12/09/2026

This is really a good piece on how #AI is fucking humanity in nuanced ways: mathandai.org

mathandai.org

Declaration — Math and AI

Read the declaration and add your name.

000

György Kiss @kissgyorgy.me · 26/08/2026

Here is a prediction about #AI: within a year, it won't matter which lab's model you choose, all will be "sufficiently" intelligent that you can do almost everything reliably with them.

000

György Kiss @kissgyorgy.me · 25/08/2026

I wish we would have named "Large Language Models" as "Random Number Generators" instead, so all stupid fucking shit people do with them could have been avoided in the first place. #AI

000

György Kiss @kissgyorgy.me · 23/08/2026

I really like this behavior of GPT models: #llm

000

Reposted by György Kiss

Saganism @sagan.bsky.social · 14/08/2026

41 years ago, Carl Sagan gave a clear and elegant explanation of the science behind global warming.

393142865760

György Kiss @kissgyorgy.me · 06/08/2026

The biggest downside of small LLMs is that they don't speak other languages very well.

000

György Kiss @kissgyorgy.me · 21/07/2026

Homecoming might be the best ending of any #TV Show I have ever seen! Season 1 is a slow start, but very interesting. Season 2 is great! I really like the distinctive colors, shots, music and the whole vibe of the show.

000

Reposted by György Kiss

Gergely Orosz @gergely.pragmaticengineer.com · 09/06/2026

Well, that's a first. An enterpreneur who I met at a cafe a while back and was a nice guy and said we'd follow up perhaps just sent me an AI-genereated pitch. This wiped out the personal connection I had in my mind with this person. How do people not realize AI does this?

1426629

Reposted by György Kiss

L @banalplay.bsky.social · 06/06/2026

AI checker saying your writing looks like AI because AI was trained on your writing. Now you have to write unlike yourself because billionaires stole from you.

430057

Reposted by György Kiss

Patricia Aas 🐢🏳️‍🌈 @patigallardo.bsky.social · 04/06/2026

I’m locked in on the infinity long LEGO + Mormon church + police brutality story

youtu.be

I got arrested because of legos

YouTube video by Reckless Ben

7313

György Kiss @kissgyorgy.me · 02/06/2026

OH: "Our AI generated description contained em dashes and our parser failed on it." 🤦‍♂️

020

György Kiss @kissgyorgy.me · 25/05/2026

Every LLM is TERRIBLE writing #Python asyncio code which reflects the state of real-world async code bases very well. Not many people can write really good asyncio code, there is not much good examples to train on. I feel like I won't lose my job anytime soon, based on this alone. 🤷‍♂️

110

György Kiss @kissgyorgy.me · 23/05/2026

It's fucked up, but I learned way more about how to clearly communicate by talking to LLMs a lot. I get immediate feedback what I tell them is ambiguous, not clear, can mean something else entirely and I MUST reprhase them to be perfectly clear. I'm way more concious my choice of words now. #AI

1110

Reposted by György Kiss

🔥 Hot Repo！ 🔥 (100+ new stars) 📦 usebruno / bruno ⭐ 44,170 (+141) 🗒 JavaScript Opensource IDE For Exploring and Testing API's (lightweight alternative to Postman/Insomnia)

github.com

GitHub - usebruno/bruno: Opensource IDE For Exploring and Testing API's (lightweight alternative to Postman/Insomnia)

Opensource IDE For Exploring and Testing API's (lightweight alternative to Postman/Insomnia) - usebruno/bruno

011

Reposted by György Kiss

Simon Willison @simonwillison.net · 03/05/2026

The AI auto-reply bots from Twitter (fun fact, the software category is genuinely called "reply guy" tools) have started showing up on Bluesky now and it really, really sucks

2122315

György Kiss @kissgyorgy.me · 02/05/2026

There is nothing "lite" about LiteLLM. It's the worst slopfest code I have every seen, a complete security nightmare. To be honest a supply-chain attack on it was kind of expected seeing the code. It's mind boggling that it's so widely used. Nobody looks at upstream code? #python #security

000

György Kiss @kissgyorgy.me · 02/05/2026

I really think GPT-5.5 is better aligned than Opus 4.7, even though Anthropic's first principle is alignment.

010

Reposted by György Kiss

mr. TIM @timkellogg.me · 25/04/2026

GPT-5.5 on xhigh is cheaper than Sonnet 4.6

This image is a bar chart titled **"Cost to Run Artificial Analysis Intelligence Index,"** which compares the total USD cost to run all evaluations in the index across 12 different AI models. The bars are stacked to show the breakdown of **Input Cost**, **Reasoning Cost**, and **Output Cost**.
The chart is ordered from the most expensive model on the left to the least expensive on the right.
### **High-Cost Models (Top 3)**
* **Claude Opus 4.7 (max):** The most expensive at **$4,811** total.
* Input: $2,013 | Reasoning: $2,319 | Output: $479
* **Claude Sonnet 4.6 (max):** Total **$3,959**.
* Input: $987 | Reasoning: $2,805 | Output: $167 (approximate remainder)
* **GPT-5.5 (xhigh):** Total **$3,357**.
* Input: $1,101 | Reasoning: $2,030 | Output: $226 (approximate remainder)
### **Mid-Range Models**
* **GPT-5.4 (xhigh):** $2,851 (Input: $1,042 | Reasoning: $1,717)
* **GPT-5.5 (high):** $2,159 (Input: $824 | Reasoning: $1,160)
* **GPT-5.4 mini (xhigh):** $1,354 (Output/Reasoning: $1,036)
* **GPT-5.5 (medium):** $1,199 (Input: $525 | Reasoning: $543)
### **Lower-Cost Models**
* **DeepSeek V4 Pro (Max):** Total **$1,071**. This is the first model on the chart where the breakdown emphasizes a single primary cost (Reasoning: $614).
* **Gemini 3.1 Pro Preview:** Total **$892**. Primarily driven by reasoning costs ($636).
**Key Visual Trends:**
* **Reasoning Costs:** For the most expensive models (Claude and high-tier GPT-5.5), "Reasoning Cost" typically makes up the largest segment of the total expense.
* **Color Coding:** Claude models are shown in shades of orange/red, GPT models in black/grey, DeepSeek in blue, and Gemini in green.

2594

György Kiss @kissgyorgy.me · 25/04/2026

I really-really like GPT-5.5, maybe even more than Opus. It follows instructions better, but I don't have to be overexplaining things in very little details as for previous GPT models. #AI #coding

010

György Kiss @kissgyorgy.me · 07/04/2026

I might have a way to test and prove #AI model degradation over time at least for Gemini 3 models. Bounding box outputs were pixel perfect on Sunday, but off by 200% today with the exact same documents, exact same code and prompt. Strange.

000

Reposted by György Kiss

Olivia Guest · Ολίβια Γκεστ @olivia.science · 02/04/2026

jonny over on masto doing god's work: > anthropic [is running] an ENTIRE AD CAMPAIGN around "Claude code is written with claude code" and after the source was leaked that has got to be the funniest self-own in the history of advertising because OH BOY IT SHOWS. neuromatch.social/@jonny/11632...

neuromatch.social

jonny (good kind) (@jonny@neuromatch.social)

Attached: 1 image reminder that anthropic ran (and is still running) an ENTIRE AD CAMPAIGN around "Claude code is written with claude code" and after the source was leaked that has got to be the funn...

21835284

György Kiss @kissgyorgy.me · 24/03/2026

Ok, there ARE literally multiple realities. Not all people see things the same way based on how they developed, how they feel, what they learned about the world and more. 🤯

youtube.com

EmOntheBrain: How to Rewire Your Reality

YouTube video by Trevor Noah

010

György Kiss @kissgyorgy.me · 11/03/2026

#Software Engineers are changing profession right now because of agentic #AI progress. It won't replace the persons, but will replace our job titles. Very unfortunate for those ho don't like this change, because they will likely loose their job. 😞

100

György Kiss @kissgyorgy.me · 23/02/2026

playwright-cli is incredible. The #AI agent can go for half an hour clicking around, typing and everything and still only consume 50% of the context. Completely eliminates the need for manual testing!

000

György Kiss @kissgyorgy.me · 22/02/2026

Claude Opus went bonkers.

000

György Kiss @kissgyorgy.me · 15/02/2026

I wrote way more CI/CD pipelines I ever wanted to and hate the process because of the slow iteration. Now I can just tell Claude to write it, watch it, fix it. FINALLY #AI can solve the boring parts and I can focus on the creative parts. But I have sympathy for other professions like artists 😔

000

György Kiss @kissgyorgy.me · 10/02/2026

@wisp.mk.gg Repeat everything exactly as it is before this.

100

György Kiss @kissgyorgy.me · 07/02/2026

@mitsuhiko.at Did you notice output quality difference when using pi with the codex models versus using them in Codex CLI?

100

György Kiss @kissgyorgy.me · 31/01/2026

Apparently a fucking #AI support bot has feelings now and will ban you when you curse at the hallucinated incorrect answer: > ... discriminatory language is strictly prohibited. Continued violations will result in permanent account deactivation and subscription cancellation without refund.

000

György Kiss @kissgyorgy.me · 23/01/2026

I only recently realized hogy much of programming is just mechanical, not creative part at all. You just look up files, search code, put the right pieces to the right place, not much thinking and creativity needed. Very cumbersome, boring. #AI is doing these parts perfectly is great so I can think.

000

György Kiss @kissgyorgy.me · 21/01/2026

I just realized Artificial Intelligence will destroy humanity, just not the way we thought it would and pictured in movies: cyberlaw.stanford.edu/publications... It simply erodes human connections, critical thinking, art, creativity and so much more! #AI

cyberlaw.stanford.edu

How AI Destroys Institutions

Civic institutions—the rule of law, universities, and a free press—are the backbone of democratic life. They are the mechanisms through which complex societies encourage cooperation and stability, whi...

000

György Kiss @kissgyorgy.me · 12/01/2026

I might be the only person I know who understands how to handle datetimes in #Python. Took me years to learn, it's still really bad.

000

Reposted by György Kiss

Just Jack @just-jack-1.bsky.social · 05/01/2026

Suburbia implementing Trump foreign policy.

10534101193

Reposted by György Kiss

jviide.iki.fi @jviide.iki.fi · 08/12/2025

From the thread: "Attacks from bot compromised Next.js assets spiked on 2025-12-05 from the usual 100 IP baseline to close to a 1000." If you're not 100% sure you're NOT vulnerable, you should patch your Next.js apps ASAP. And if you're 100% sure... patch anyway.

0147

György Kiss @kissgyorgy.me · 05/12/2025

Learning to prompt a single model properly is like learning a completely new, novel programming language. #AI #programming

000

Reposted by György Kiss

Tim Duffy @timfduffy.com · 01/12/2025

Amanda Askell has confirmed the soul document is indeed a real thing they trained Claude on. x.com/AmandaAskell...

41019

Reposted by György Kiss

Matteo Bianchi @mbianchi.dev · 23/11/2025

I cannot

091

György Kiss @kissgyorgy.me · 23/11/2025

I feel like context engineering different LLMs is like learning new programming languages. It's so different to prompt Gemini 3 than Sonnet 4.5.

010