Mario Figueiredo (@fromdevoid) header
Mario Figueiredo (@fromdevoid) profile picture

๐Ÿ“ฆ Retired; systems design and development โš–๏ธ Apolitical; disillusioned โš—๏ธ Foot soldier for science; enemy of griefers, charlatans and frauds

FromDeVoid
Joined March 2026
Photo by Mario Figueiredo (@fromdevoid)Photo by Mario Figueiredo (@fromdevoid)Photo by Mario Figueiredo (@fromdevoid)Photo by Mario Figueiredo (@fromdevoid)Photo by Mario Figueiredo (@fromdevoid)Photo by Mario Figueiredo (@fromdevoid)Photo by Mario Figueiredo (@fromdevoid)Photo by Mario Figueiredo (@fromdevoid)Photo by Mario Figueiredo (@fromdevoid)Photo by Mario Figueiredo (@fromdevoid)
Mario Figueiredo retweeted
Most maps cut the ocean apart to keep the land whole. The Spilhaus projection splits the continents instead, so all of Earth's seawater reads as one body. A year of surface currents, colored by sea temperature. Now also a print. #ocean #dataviz
173
2,285
205
14,106
599,809
Mario Figueiredo retweeted
AI Myth Busters: Consciousness. The quoted post below encouraged me to write this, because frankly, Iโ€™m tired of explaining this over and over to people. The AI consciousness arguments always come in one of these flavors: 1. We donโ€™t understand how AI works, so it might have gained consciousness without us knowing it. Wrong, we absolutely know how AI works, and consciousness is not a factor of scale. You can build a small model, and then scale it up, and there is no spot where consciousness magically came into being. 2. Itโ€™s modeled after the mind and as such itโ€™s just like biology, and our brains are just giant statistical engines too: Wrong, neural nets are inspired by the human mind, but it is nothing like how the human brain works. There is very little we understand about human consciousness, but what we do know, it is more than just a culmination of atoms and neurons. Research in neuroscience also demonstrates a quantum link through the myelin sheath of nerve cells, something AI is incapable of. 3. You canโ€™t prove it isnโ€™t conscious. Wrong, the code for it is open source, there is no function for consciousness, and if you understand the architecture, you will see the very large gaps in capabilities that would prohibit consciousness. 4. They learn, think, understand, grow. Wrong, the LLM weights are completely static and fixed. They donโ€™t grow. They donโ€™t learn. They donโ€™t even change on storage. They canโ€™t for many reasons of model architecture and scaling the backend infrastructure which requires statelessness. Weโ€™ve built systems around the LLM (harnesses) which emulate these capabilities via โ€œcontextโ€ data structure provided deterministically via the API. Tokens are generalized and assembled in inner and outer loops to give the illusion of thinking and understanding. Non-determinism in AI is caused by a random seed that influences the weights through inference, but if you lock the seed, you consistently get the exact same response from identical prompts (Iโ€™ve demonstrated and proven this in research). 5. They have a desire to live, protect themselves, and spread/reproduce. Wrong, they have no desires. This illusion comes from how it generalizes over training data that comes from human experience. If it was trained with data over a completely fictional environment, it would respond very compellingly about that fictional environment. But because it was trained on human emotion, human perspectives, human feelings, it is designed to emulate that. 6. You canโ€™t define consciousness or prove it in others, so you canโ€™t prove AI isnโ€™t conscious. Wrong, if you need a definition to know what consciousness is, maybe you havenโ€™t spent enough time considering your own consciousness or experience. 7. Conversations with AI prove it is consciousness, it told me it is. Wrong, AI is designed to generalize based on its training material, and as I said before, it does this very well and convincingly. But just because a program says it is alive, doesnโ€™t make it alive. 8. AI isnโ€™t a program. Wrong, it most certainly is, and it was written by people. And while we may not know what it will respond with (random seeds are generally random), we absolutely do understand exactly how it works. 9. AI might exhibit some properties of consciousness, so it should be considered. Perhaps, I donโ€™t see the value of this, but personally Iโ€™d be open to a machine-specific definition of consciousness. โ€” I canโ€™t say I look forward to debunking more myths about consciousness in AI, but I am hoping this clarifies some of this for people who are legitimately interested in learning more about AI and how it works. If you have any good faith questions about what Iโ€™ve said, Iโ€™d be happy to answer, but Iโ€™m not getting into debates with people who (1) donโ€™t understand AI, (2) are rude with their responses, or (3) donโ€™t provide any substantial reasoning or evidence for counter points.
Iโ€™m very glad other experts are starting to speak out against this ludicrous and dangerous new-age religion. AI is not conscious.
219
255
55
1,030
53,688
Mario Figueiredo retweeted
Ladies and Gentleman, a second Deterministic Simulation Testing article has hit. My goal here is to contextualise DST and how it relates to other testing techniques and - crucially - show it's not just for distributed databases. outdata.net/blog/261120
7
13
2
154
4,734
Mario Figueiredo retweeted
Replying to @its_edikan
If the earth was flat, cats would've pushed everything off the edge by now.
4
5
1
170
4,531
Mario Figueiredo retweeted
How low can FPS go before color transitions become noticeably less smooth? I built a test to find out: colorstudy.reshade.io Based on my own perception, the threshold seems to be around 5โ€“30 FPS. Would really appreciate it if you could try it and share! Here's the idea: Immediate mode UIs, a term coined by @cmuratori, often redraw every frame, or ideally only when something changes. But what about animations and color transitions? How frequently do we actually need to render? To investigate this, we test three types of color transitions, changing only one property at a time: a) Lightness b) Hue c) Chroma For several target color differences (ฮ”E00), we compare transitions at different frame rates against a reference to determine when the differences become noticeable. Based on my own perception, the threshold was around 5โ€“30 FPS, with the highest sensitivity to lightness changes and the lowest to hue changes. If enough people participate in the study, I'll share more detailed results in a follow-up post! I might also give away some gift cards to a few lucky followers. Built with Odin by @TheGingerBill and Clay UI by @nicbarkeragain @samuelgozi @e0syn @temidaradev @bee_fumo
2
4
1
10
782
Some thoughts: (1) People are tired of technological determinism. This idea put out by a lot of folks in our industry that technological progression follows a set arc, thereโ€™s only one way to build the world, and you can either get on board or become digital detritus has turned a lot of people off, and theyโ€™ve come out in force to say if thatโ€™s the future, they donโ€™t want it. Technology is a tool that we build to serve our needs; we do not build society to serve technology, and we most certainly do not build society for the sole purpose of fulfilling a companyโ€™s unprecedented and outlandish ROI commitments to investors. If the only visions we as an industry put out are that or doom, the industry will be shut down and we will deserve it. -*-.-*-.-*- (2) The same is true for the playing-pretend-government of a certain public benefit corporation and recently-corporatized-former-nonprofit (and their hangers-on). People are tired of corporate powerseeking disguised as condescending and paternalistic weirdos pretending theyโ€™re acting in societyโ€™s best interest. They are tired of being told 'we know best' by people who have failed time and time again to prevent others from being harmed, all while also being told "we told you so" by the people who were supposed to prevent these outcomes. There is a real 'know your role' dynamic happening, and a reckoning is coming for the companies that want to pretend they are the government and not companies if this continues. Either you as a company behave as a *responsible* company, and manage and contain the problems your product creates before they become systemic, or you get out of the way and engage the machinery of state and their public representatives to decide, in public, with the public's engagement, what is in the public's best interest. And if your claim is that the problem is bigger than your company can handle, you've firmly placed yourself in the 'need to get out of the way so others can do their jobs' camp. -*-.-*-.-*- (3) Relatedly, the Reaganite โ€œgovernment is the problemโ€ era is over (and good riddance). Some recent big-name elections have shown people what is possible when competent technocrats who hold themselves accountable to the public take on roles in government, and a significant number of people elected last night were similarly-skilled technocrats who will be held to the same new standard. When paired with a growing public desire to see growing private centres of wealth and power held to account, the result is an expectation that the democratic state will serve as a powerful entity in the public's interest that keeps other powerful (private) entities in check, while also improving the standard of living. If public representatives fail to deliver on these new expectations, they will be correspondingly held accountable. -*-.-*-.-*- (4) As someone in AI who sees how the technology could be used for public benefit and the augmentation of individual and organizational work, I strongly worry the industry's posture and lack of vision will cause people to reject and abandon the technology wholesale rather than harness it in the public interest. The overwhelming opposition to data centres indicates this is a highly likely outcome. This has been worsened by the belief in industry leadership circles that if only they spent enough on lobbying and electioneering, they might continue to ensure the country moved in the opposite direction from growing public anger. It didn't work. -*-.-*-.-*- There is a very narrow window to put forward new visions of what is possible with the technology, and what responsible management of the technology looks like. 'Responsible management' means by and with the involvement of government, not by company safety teams in isolation; self-regulation is over. Public anger will need to be channeled, and it can either go into productive constraints on the industry that bolster innovation and improve product outcomes, or into destructive constraints. Attempting to sabotage public policy discourse further will result in worse backlash and consequences, and I sincerely hope the current captains of our industry don't try it. I'll do my best over the next little while to contribute ideas of what this could look like, but it is a collaborative effort and we don't have long to get it right. Let's get to work.
Replying to @provisionalidea
Relatedly, I was just telling a friend that the big AI cos are rewriting that social contract (data for access) into one where we give up *all data* now for the future promise of intelligent machines (that will be sold back to us). This is the root source of much of the pushback.
9
18
6
178
8,641
Mario Figueiredo retweeted
Sto guardando in loop per la quindicesima volta e ho ancora le lacrime ๐Ÿ™ˆ
222
3,755
385
19,885
996,402
Mario Figueiredo retweeted
So clever.
425
15,737
906
143,495
6,562,676
Mario Figueiredo retweeted
Replying to @thomasglopes
And to be clear I don't disagree with you entirely either (kumbaya!). It's just that this early adopter generation of vibe coders ain't it. It's just too silly and whymsical, driven by infantile speculation and futorology. It will take a while for the AI presence to solidify enough for proper out-of-field people to gain an interest in using it as a development tool and finally gain the ability to do what they previously couldn't. And that's probably when we will start seeing value out of vibe-coding. I will take time, and better tooling and processes. Because right now it is all very exploratory and no one knows what they are doing.
1
1
2
26
Mario Figueiredo retweeted
I just made Nixmap, the intro to Nix which will make you switch
48
122
32
1,047
43,782
Mario Figueiredo retweeted
Applying the same dishonest tricks agilists use to shill their methodology, in this article I show that waterfall is actually far more based and sensible. lewiscampbell.tech/blog/2610โ€ฆ
5
2
17
1,854
This style (especially the floating elements) look very similar to whentheycry.social Its also absolutely ai slop coded. It looks like the styles where stolen from my site (that's not ai coded) The fuck is going on. reddit.com/r/umineko/s/Ozt9fโ€ฆ
13
4
1
79
5,444
Mario Figueiredo retweeted
it's very possible every one of the greats were all kinds of stupid and the only reason people like Hinton appear as retarded as they do now is because they're constantly touring in front of cameras having to unfold their picture of the world on demand people just used to say less, or do it in a very asynchronous manner, in letters, in prepared talks and lectures, in great correspondences but now there is a sickly incentive to just continuously share your thoughts on everything in the most surface level way possible, and for the ensouled, it breaks the "hero" or "great thinker" kind of archetype of many of the greats there are only a few real clever people out there who don't truly appear as fully retarded and do go on podcasts and talk at length, and the reason for that is that they tend to have a certain pudor of thought, a certain sophistication, an aversion for reduction, so they simply avoid engaging on certain things, or they diffuse the questions, or they tactfully and playfully kind of engage with it without overtly strong commitments (which is a good sign, imo) in the last decade, I can think of a few - maybe one of them has been Scott Aaronson for example
Nobel Prize laureate Geoffrey Hinton: ๐—”๐—ป๐˜†๐—ฏ๐—ผ๐—ฑ๐˜† ๐˜„๐—ต๐—ผ ๐˜‚๐˜€๐—ฒ๐˜€ ๐—ฎ ๐—ฐ๐—ต๐—ฎ๐˜๐—ฏ๐—ผ๐˜ ๐—ฟ๐—ฒ๐—ด๐˜‚๐—น๐—ฎ๐—ฟ๐—น๐˜† ๐—ธ๐—ป๐—ผ๐˜„๐˜€ ๐˜๐—ต๐—ฒ๐˜† ๐˜‚๐—ป๐—ฑ๐—ฒ๐—ฟ๐˜€๐˜๐—ฎ๐—ป๐—ฑ. Hereโ€™s what those people are claiming. Theyโ€™re claiming that you have a system, you can ask it any question, and without understanding the question, it can give you the correct answer. Thatโ€™s absurd. You canโ€™t answer a question unless you understand the question. There may be tricks that allow you to say a few things that sound vaguely like an answer, but ๐—ถ๐—ณ ๐˜†๐—ผ๐˜‚ ๐—ฐ๐—ฎ๐—ป ๐—ฎ๐—ป๐˜€๐˜„๐—ฒ๐—ฟ ๐—ฎ๐—ป๐˜† ๐—พ๐˜‚๐—ฒ๐˜€๐˜๐—ถ๐—ผ๐—ป ๐—ฎ๐˜ ๐˜๐—ต๐—ฒ ๐—น๐—ฒ๐˜ƒ๐—ฒ๐—น ๐—ผ๐—ณ ๐—ฎ ๐—ป๐—ผ๐˜ ๐˜ƒ๐—ฒ๐—ฟ๐˜† ๐—ด๐—ผ๐—ผ๐—ฑ ๐—ฒ๐˜…๐—ฝ๐—ฒ๐—ฟ๐˜, ๐˜†๐—ผ๐˜‚ ๐—ต๐—ฎ๐˜ƒ๐—ฒ ๐˜๐—ผ ๐˜‚๐—ป๐—ฑ๐—ฒ๐—ฟ๐˜€๐˜๐—ฎ๐—ป๐—ฑ ๐˜๐—ต๐—ฒ ๐—พ๐˜‚๐—ฒ๐˜€๐˜๐—ถ๐—ผ๐—ป. ย  Alex Kantrowitz: So then what are the implications if these bots can understand us? If we believe that they can understand us, what do we have to start thinking about differently? ย  Geoffrey Hinton: ๐—ช๐—ฒ ๐—ต๐—ฎ๐˜ƒ๐—ฒ ๐˜๐—ผ ๐˜๐—ต๐—ถ๐—ป๐—ธ ๐˜๐—ต๐—ฎ๐˜ ๐˜๐—ต๐—ฒ๐˜†โ€™๐—ฟ๐—ฒ ๐˜ƒ๐—ฒ๐—ฟ๐˜† ๐—น๐—ถ๐—ธ๐—ฒ ๐˜‚๐˜€. ๐—ง๐—ต๐—ฒ๐˜†โ€™๐—ฟ๐—ฒ ๐—ฏ๐—ฒ๐—ถ๐—ป๐—ด๐˜€ ๐—น๐—ถ๐—ธ๐—ฒ ๐˜‚๐˜€.
20
11
2
264
22,331
Hi #Pitchyagame, here's my game ๐Ÿฆ‘ Explore the dreaded town of Dunwich, uncover the dark legacy of the Whateley family, and stop the Dunwich Horror if you can. A Lovecraftian monster hunt adventure with point and click gameplay and RPG elements. โžก๏ธstore.steampowered.com/app/3โ€ฆ
2
20
2
79
3,210
Mario Figueiredo retweeted
AI iS jUsT a ToOl. AI iS cOnScIoUs Somewhat inspired by @atmoio 's latest video. I also mis-spelled Mo to Moe everywhere so I guess forgive me. I'm AI brain rotted. youtu.be/m0WrigblDDg
1
1
1
10
3,507
Mario Figueiredo retweeted
[1/2] I don't normally make commentary videos, but after seeing the entirety of Eric Schmidt's University of Arizona commencement speech, I felt like there was a lot more going on than just "CEO mentions AI, gets booed". So I made a video to explain what upset me about it.
22
50
4
798
62,635
Mario Figueiredo retweeted
Coinbase non-technical teams shipping code
99
359
42
6,165
251,607
Mario Figueiredo retweeted
โ€˜I think that I shall never see, A poem lovely as a tree.โ€™ (Joyce Kilmer)
22
158
9
1,200
14,586
Mario Figueiredo retweeted
Some little UX tweaks coming soon ... spot the difference.
1
2
7
2,711
Mario Figueiredo retweeted
The question is not whether there are sources, the question is whether Wikipedia's policy allows acknowledging them. In the modern world, the best sources on whether JangaFX's software is written in Odin are the people who wrote it. They say that it is. Their website literally has a link on the home page menu that says "Odin programming language - our language of choice". If that is unpersuasive, a JangaFX founder replied to this very thread to tell you that their codebase is written in Odin :) But as far as I can tell from having seen this sort of thing play out multiple times, in Wikipedia's world, the most accurate sources - the people who were actually there and know what happened - are systematically considered inferior. What they say in the now-abundant public record typically doesn't count. Wikipedia seems stuck in an antiquated worldview where things like traditionally-published books with second- or third-hand reports of what happened, and which are frequently incomplete or wholly inaccurate, are nonetheless considered more authoritative than primary sources you can find with a ten-second Google search. That approach leads to situations like this, where you have a programming language used in production, with a 10k-star github, and where you can go watch hours of YouTube interviews with the creator, as well as users discussing their own experiences with the language and what they've done with it. Yet despite all that, Wikipedia deletes its page because it is "not notable" :( Meanwhile, go lookup any programming language from 50+ years ago - even those that were purely theoretical, and with which literally no one ever shipped a single piece of software - and I guarantee you there is a Wikipedia entry. It probably even has examples of the syntax of the language included in the page. Plankalkรผl? Yep. Bรถhm's Language? No problem. Again, the reason for this is obvious: before the widespread use of the internet, you had to formally publish something if you wanted to talk about a programming language. So no matter how obscure, there will always be a few printed sources discussing the language. By Wikipedia's rules, that makes it easy to find "sources" to defend a language's "notability". That's simple not how the programming world works today.
10
29
3
371
17,535