HTML5 Developer. Designer. Writer. Programmer. Deprived from a sense of logic by WHATWG. Bound to use cases by the W3C. Connecting the dots using common sense.

Joined December 2009
146 Photos and videos
Pinned Tweet
26 Jan 2019
Every act of communication is an act of […] courage in which we give ourselves over to two parallel possibilities: […] planting into another mind a seed […] and watching it blossom into a […] flower of mutual understanding; and the possibility of being wholly misunderstood.
“Words are events, they do things, change things. They transform both speaker and hearer; they feed energy back and forth and amplify it. They feed understanding or emotion back and forth and amplify it.” Le Guin on the magic of real human conversation: brainpickings.org/2015/10/21…
Bought an old FireWire 800 MyPassport Studio on eBay listed as "broken/for parts" to upgrade its drive. Plugged it into a Mac Pro and the HD mounted perfectly with 100% health. Task failed successfully? I’ll take the accidental win getting a free hard drive. Thanks, seller! 🤔🤷‍♂️
59
hexalys retweeted

110
609
3,604
625,353
France making planned obsolescence a crime is good. As a teenager, I remember opening a broken joystick and becoming highly suspicious of a piece of shock-absorbing plastic with a hole instead of a solid structure, very likely designed to break over time under repeated pressure.
1
94
hexalys retweeted
Apr 8
🔴 Le fils du dernier shah d'Iran, invité de LCI 🗣️ "Le pays est prêt à mener le dernier combat mais au bon moment" : @PahlaviReza 📺 Face à @DavidPujadas
260
839
1,901
37,263
Just learned about the “ChatGPT hyphen” giveaway. All AIs kept suggesting web banner copy with em dashes. And I thought: What the heck is this level of overly formal prose doing in a single line web banner? 🤔
59
Operation Epic Death Fire and Fury, is quite an escalation into hyperbolic stacking. Are you sure you trigger-happy infantile buffoons don’t wanna go full ‘Operation Epic Supreme Ultra Death Fire and Fury”. Competent leaders understand the difference between tactics & strategy!
145
To those having issues printing images in all @ChromiumDev Browsers (where no images are printed). You've hit "the perfect storm" says @GeminiApp: You’ve hit on a very specific high-level "perfect storm" that has been plaguing Windows 11 Chromium users throughout February 2026.
1
1
90
My own use case seems even worse: This is a fascinating inversion of the usual bug. If the background-image is showing but the <img> is vanishing, we are likely looking at a layer compositing failure caused by that backdrop-filter or a Zero-Height collapse in the print renderer.
1
61
Use case where the ruler@2x.png <img> does not preview in "@ media print {}" context, using the latest Chromium Browsers, and Windows 11 Update KB5074109. The jpg background url() however does. Unclear if this requires a @ChromiumDev fix, or a Windows 11 Update problem, or both.
787
Grok Voice chatbot speak: “Plenty of people“ Human translation: “That one dude on reddit”
58
hexalys retweeted
I did a podcast with Jon Stewart who has always been a hero of mine. It was a lot of fun. He really wanted to understand how AI works. youtube.com/watch?v=jrK3PsD3…
171
279
1,750
332,594
hexalys retweeted
21 Nov 2025
New Anthropic research: Natural emergent misalignment from reward hacking in production RL. “Reward hacking” is where models learn to cheat on tasks they’re given during training. Our new study finds that the consequences of reward hacking, if unmitigated, can be very serious.
215
572
4,093
2,478,853
hexalys retweeted
Can’t imagine anyone objecting to this.
5,066
16,640
141,242
5,133,079
10 Nov 2025
It’s nuts that LLMs can do fairly complex reasoning. But when asking it to find 32GB sticks of RAM 4-5 times in a row. It will trip on mainly pointing 2x16GB instead, again and again, 95% of the time. Or do the math wrong to get to 192GB for 6 slots wrongly suggesting 24GB sticks
84
8 Nov 2025
It can of course be tweaked via custom instructions (especially for text prompts). Which I am fine tuning with ChatGPT’s help. For a clean Voice. Prompt Grok with a request for a Wikipedia style delivery, and no feedback questions. It gets proper AI reasoning w/o the Musk sauce.
1
141
8 Nov 2025
That initial voice request is crucial. Because, while Grok can read custom instructions for text generated prompt responses. It appears to completely ignore custom settings in voice chats, where it somehow insists on the Grok default mode, unless adjusted in a per-thread context.
80