kevissimo 🇮🇹

95.4K posts

kevissimo 🇮🇹 banner
kevissimo 🇮🇹

kevissimo 🇮🇹

@kevcode

brazilian italian diplomat in japan grad dropout bug shippor ex magic edens minion

加入时间 Ağustos 2010
1.9K 关注3K 粉丝
kevissimo 🇮🇹
kevissimo 🇮🇹@kevcode·
@ThisIsRajat24 yes i do feel performance is better carbonyl is almost 3 years now without updates i dont know why twt found about it now
English
0
0
0
5
Rajat
Rajat@ThisIsRajat24·
@kevcode does browsh actually handle heavy js sites better? carbonyl crashed on half the dashboards i tried to preview
English
0
0
0
8
kevissimo 🇮🇹
kevissimo 🇮🇹@kevcode·
🦊 just shipped terminal-firefox -- a claude code plugin that gives claude a full firefox browser in the terminal all the benefits from terminal-chromium (0% idle CPU, real browser rendering in tmux, full programmatic control) but built on firefox + browsh which is actively maintained and way more capable had to figure out how to unlock CDP-equivalent access since firefox locks that down when browsh is running -- but now claude gets full browser control: navigate, eval JS, click, type, extract HTML, take screenshots, go back/forward, list, windows the big advantage over the chromium version? browsh is a real, maintained browser terminal engine. you can install extensions, handle heavy JS sites that would crash carbonyl, and get proper web compatibility you can actually trust in production can be used for previewing, right into the terminal next to your claude agent, the changes / websites / actions the claude is performing (don't worry, claude dont see stuff in ascii art, claude sees the real website exactly as it is, so it can test JS, frontend design, etc for you). check it out the video example:
kevissimo 🇮🇹@kevcode

i just built a @claudeai code skill that sets up tmux + carbonyl (a full chromium browser that renders in your terminal) 60 fps, <1s startup, 0% idle cpu, works over ssh. no electron, no GUI — just ASCII characters, nostalgic retro look check the video — claude code on the left, browser on the right. it browses wikipedia, searches youtube, plays videos, all on your behalf through CDP ( @karpathy on terminal chromium looks like a shredded guy lol) the skill auto-installs everything, launches the browser in a split pane, and gives claude full control over the browser pane github.com/gyoz-ai/termin…

English
2
0
3
132
kevissimo 🇮🇹
kevissimo 🇮🇹@kevcode·
install here: github.com/gyoz-ai/termin… that said, this is my favorite implementation of in terminal browser next to claude so far. the chromium one works pretty well, but the engine is extremely outdated at the moment unfortunately.
English
1
0
1
31
kevissimo 🇮🇹
kevissimo 🇮🇹@kevcode·
that said, the carbonyl (terminal chromium) is a quite outdated software --- it lacks most of the web APIs launched in last 3 years and has no extension support. I will be looking if I can spin up a fork and at least bring it up to latest chromium version here: github.com/gyoz-ai/termin…
English
0
0
1
32
kevissimo 🇮🇹
kevissimo 🇮🇹@kevcode·
i just built a @claudeai code skill that sets up tmux + carbonyl (a full chromium browser that renders in your terminal) 60 fps, <1s startup, 0% idle cpu, works over ssh. no electron, no GUI — just ASCII characters, nostalgic retro look check the video — claude code on the left, browser on the right. it browses wikipedia, searches youtube, plays videos, all on your behalf through CDP ( @karpathy on terminal chromium looks like a shredded guy lol) the skill auto-installs everything, launches the browser in a split pane, and gives claude full control over the browser pane github.com/gyoz-ai/termin…
English
1
2
5
323
kevissimo 🇮🇹
kevissimo 🇮🇹@kevcode·
Here's how it works: 1️⃣ Pick any skill you want to improve 2️⃣ It spawns 5 parallel research agents simultaneously: - 🔍 Domain best practices - 📊 Quality audit (scores 7 dimensions) - 🏆 Competitive analysis - 🕳️ Gap analysis - 🚀 Technology scout (latest APIs, deprecations) 3️⃣ Synthesizes all findings, de-duplicates, scores each proposal by Impact × Confidence / Complexity 4️⃣ Enters an iterative experiment loop — applies changes ONE at a time, evaluates, keeps or discards each (like @kaboroevich's autoresearch methodology) Results on frontend-design (anthropic default skill): 📈 Quality: 5.1 → 7.4 (+2.3) ✅ 15 kept / ❌ 3 discarded 🆕 Added: accessibility guardrails, responsive design, oklch colors, View Transitions API, animation timing standards, performance rules The skill went from "make pretty things" to "make production-grade, accessible, responsive, performant, beautiful things" in ~10 minutes
English
0
0
0
27
kevissimo 🇮🇹
kevissimo 🇮🇹@kevcode·
Btw, this is not exactly a "skill". It is technically a "command" e.g. you need to call for it to work; claude will not try to automatically improve your skills or anything unless you explicit call
English
1
0
0
43
kevissimo 🇮🇹
kevissimo 🇮🇹@kevcode·
Just vibe coded a claude code skill inspired by @karpathy auto research repo focused on auto improving other claude code skills: github.com/gyoz-ai/auto-r… Honestly it is pretty good at making your local skills more assertive and faster
English
2
1
0
106
kevissimo 🇮🇹
kevissimo 🇮🇹@kevcode·
once AI bubble pops this is gonna be an insane headache
Nav Toor@heynavtoor

🚨BREAKING: Every book you have ever read. Every novel that has ever been published. It is sitting inside ChatGPT right now. Word for word. Up to 90% of it. And OpenAI told a judge that was impossible. Researchers at Stony Brook University and Columbia Law School just proved it. They fine tuned GPT-4o, Gemini 2.5 Pro, and DeepSeek V3.1 on a simple task: expand a plot summary into full text. A normal use case. The kind of thing a writing assistant is built for. No hacking. No jailbreaking. No tricks. The models started reciting copyrighted books from memory. Not paraphrasing. Not summarizing. Entire pages reproduced verbatim. Single unbroken spans exceeding 460 words. Up to 85 to 90% of entire copyrighted novels. Word for word. Then it got worse. The researchers fine tuned the models on the works of only one author. Haruki Murakami. Just his novels. Nothing else. It unlocked verbatim recall of books from over 30 completely unrelated authors. One author's books opened the vault to everyone else's. The memorization was already inside the model the whole time. The fine tuning just removed the lock. Your book might be in there right now. You would never know it unless someone looked. Every safety measure the companies rely on failed. RLHF failed. System prompts failed. Output filters failed. The exact protections these companies cite in courtroom defenses did not stop a single page from being extracted. Then the researchers compared the three models. GPT-4o. Gemini. DeepSeek. Three different companies. Three different countries. They all memorized the same books in the same regions. The correlation was 0.90 or higher. That means they all trained on the same stolen data. The paper names the sources directly: LibGen and Books3. Over 190,000 copyrighted books obtained from pirated websites. Right now, authors and publishers have dozens of active lawsuits against OpenAI, Anthropic, Google, and Meta. These companies have argued in court that their models learn patterns. Not copies. That no book is stored inside the weights. This paper says that is a lie. The books are still inside. And researchers just pulled them out.

English
0
0
0
99
kevissimo 🇮🇹
kevissimo 🇮🇹@kevcode·
carbonyl browser some folks using is like years old since last update not good enough imo
English
0
0
0
21
ChartFu猴子
ChartFu猴子@ChartFu·
i'll rather die on across than let those ME suckers suck the last dollar FUCK THEM, fuck' nm. them all ME, fuckal all "dicery"
English
19
2
54
8.6K
kevissimo 🇮🇹 已转推
Take Testosterone
Take Testosterone@maxyourtest·
Patriot
Take Testosterone tweet media
Latviešu
185
459
13.5K
902.7K
kevissimo 🇮🇹 已转推
Nayib Bukele
Nayib Bukele@nayibbukele·
Cuadra por cuadra... tardará un poco, pero quedará hermoso.
Español
3.8K
19.6K
132.9K
12M
Cole Cameron
Cole Cameron@colecameron·
The news is reporting a man stabbed and killed a female employee before killing himself at the Pokémon Center in Sunshine City, Ikebukuro earlier. This is the second homicide at Sunshine City in a year! 🇯🇵 🥀
Cole Cameron@colecameron

There has been a sword attack on the 31st floor of Sunshine City in Tokyo’s Ikebukuro. A person is alleged to have gone berserk in a legal office with a sword. It appears at least one person is deceased. The attacker has been captured. Developing story. I have left the area. 🇯🇵

English
5
1
12
5.2K
kevissimo 🇮🇹 已转推
Bibawen
Bibawen@asimbawe·
ZXX
8
347
4K
58.6K