Cognitive Bias

Adelaide, South Australia
Joined December 2008
So, @X there is no new plan that's on par with the existing Premium+, it's either degrade, or upgrade for extra $100 per year.
1
40
Your email says it should be cheaper, not more expensive.
6
Not sure if that is intended by the @spacexai team. I thought Grok was the only non woke one. But turns out not entirely - refuses to search Internet, what otherwise would be absolutely doable manually.
4
95
Может действительно старение скоро скоро отменят как результат быстрого накопления множества маленьких побед в самых разных областях? Вот очередное на тему
This is very, *VERY* big. A method to actually *reverse* AGE CML damage that has been considered irreversible until now. The engineered enzyme CMLase, developed with Calico, acts like a “lawnmower” to remove abundant advanced glycation end-products (such as CML) from long-lived proteins. This restores original lysine residues in ex vivo human aorta, skin, and lens tissues from elderly donors. Literally rewinding the clock on age-related tissue changes and damage back to a decades more youthful condition. And the method appears to be able to be applied to many different areas of human aging. ChatGPT agrees that it worked on AGE CML that had accumulated naturally for decades in human tissue: In the human lens: a 45% reduction measured directly by LC-MS/MS; 78% by antibody assay. 75-year-old aorta and arterial tissue: more than a 70% reduction and reversal of damage. Elderly skin: more than 55% reduction, bringing the measured CML staining *below that found in 31-year-old skin*. That last result borders on astonishing: the method chemically rejuvenated old human skin tissue *past* the level of considerably younger tissue. And this is a methodology of directed evolution of an enzyme tailored to solve a specific problem and reverse it. Like CRISPR, this new tool is something that will have many uses. True age reversal. Quite a breakthrough indeed. 75 year old heart, aorta, and skin cells reversed to be younger than tissues of a 31 year old? I’m 76. I’m onboard. This could be available through medical tourism in Panama and Thailand in 5 to 7 years. US and other more restrictive environments in 7 to 10 years. As @rand_longevity often says, “just stay alive” for the next 10 years. Then decide what you want to do with your second life. Best. Timeline. Possible. 🚀💫✨ e/acc bio/acc
4
141
Would love to see more talented people apply their talents to celebrate a country's birthday. I'm a Ukrainian Australian, but I feel proud for America when watching this.
3
60
Knowing this, it would be illogical to get excited about AI generating thousands of lines of code that is barely-to-no audited.
Someone figured out how to print free money. And to this day, no one can say for sure whether anyone used it
1
4
191
Think in atoms, not in precedents.
"Elon arrivera jamais à faire des usines de puces." Je vois ce take dix fois par jour. C'est l'erreur de raisonnement la plus prévisible qui existe, et elle se démonte en deux minutes. Les sceptiques raisonnent par analogie. Jamais par premiers principes. Leur argument complet : personne ne l'a fait en moins de 40 ans, TSMC a brûlé 165 milliards en Arizona, une fab 2nm coûte 28 milliards et prend 38 mois rien qu'à construire, Tesla a zéro expérience des semis. Donc impossible. Tout est vrai. Et tout est non pertinent. "Ça a toujours pris 40 ans" n'est pas une loi de la physique. C'est juste la vitesse d'une industrie ossifiée qui n'a jamais été attaquée par quelqu'un qui pense aux atomes plutôt qu'au précédent. Alors on décompose la chaîne, du sable au chip qui calcule. Matière première (sable, silicium, wafers) : zéro contrainte physique, juste du capital et de l'énergie. Design : Tesla le fait déjà (FSD, Dojo). Lithographie EUV : le seul vrai goulot, mais c'est un problème de chèque et de file d'attente, pas de faisabilité. Process et yield : le vrai moat. Packaging, mémoire, test : coordination et capacité, pas de physique fondamentale. Question de premier principe sur chaque ligne : contrainte dure, ou contrainte molle ? Réponse : un seul goulot dur. Tout le reste est mou. Capital, coordination, yield, aversion au risque. C'est-à-dire, mot pour mot, la maison d'Elon. Et voilà le coup que tout le monde rate. TSMC est lent parce que c'est une foundry généraliste : 305 process différents, 12 000 produits, 500 clients. Chaque variante détruit le yield. Terafab fait l'inverse exact. Une fab, un seul design de puce. Le problème combinatoire qui rend le yield infernal s'effondre. Ce n'est pas TSMC qu'il reconstruit, c'est In-N-Out face à un restaurant de 300 plats. Ajoutez l'IA. La courbe de yield, ce truc qui prenait des années d'essais empiriques, c'est un problème d'optimisation et de détection de défauts. Exactement ce que l'IA compresse. Et qui a la flotte de GPU pour le faire ? Le mec qui construit la fab justement pour nourrir son IA. La boucle se referme sur elle-même. Et le edge personnel, ce n'est pas du folklore. SpaceX lui a appris à intégrer verticalement la supply chain la plus "impossible" qui existe et à diviser les coûts par 10. Tesla lui a appris le scaling manufacturing. Personne d'autre sur Terre n'a les deux cicatrices en même temps. "Aucune expérience des semis." Il n'avait aucune expérience des fusées en 2002, aucune de l'auto en 2008. Le pattern est toujours identique : il entre dans une industrie mature qui le prend de haut, il la décompose, il intègre, il gagne. "Qui va licencier l'IP de process ?" Intel a rejoint Terafab avec son node 18A. Le moat de 40 ans, il l'achète en un partenariat. "Trop de capital." 55 milliards en phase 1, jusqu'à 119 au total. Et la demande est captive (Tesla, SpaceX, xAI), pas un pari spéculatif externe. "Ses timelines slippent." Oui. C'est la seule concession honnête. Mais c'est un argument sur le quand, jamais sur le si. SpaceX était en retard. Tesla aussi. Les deux ont gagné. En retard n'a jamais voulu dire échec chez lui. Décompose la chaîne : pas une seule impossibilité physique. Que du capital, de la coordination et du yield. Les PNJ raisonnent par précédent. Le précédent vient de se faire annuler par premiers principes. Comme d'hab.
3
116
I love LeCun's ability to keep engaging with talented and interesting people here at X despite personal conflict with the X's owner on a deep conceptual level. For the benefit of millions of readers. True researcher spirit.
I do not write posts on X. I tweet links to posts on other platforms. I like and retweet (occasionally) I comment on friends' tweets (rarely) Follow me on...⬇️⬇️⬇️
1
181
Something we've been working on...
270
704
160
9,847
1,283,005
"free roaming cow with sea view" love it.
So we cook almost completely at home, locally farmed free roaming cow and chicken with sea view and local farm vegetables and mostly local fruit Imported Brazilian premium coffee beans from Sul de Minas, ground ourselves and then pour over black coffee with stainless steel Hario V60 in ceramic mugs Glass or wooden containers No plastics anywhere When I fly I think next time I'll just make slices of steak with rice and vegetables because it makes me feel better than whatever slop they serve me on the plane even in business class! I agree food outside is mostly contaminated microplastics soy slop now
3
180
This Mother’s Day, celebrate Mum — the heart of everything. Give her a feminine flamenco painting that honours her strength, grace and endless love. Original wet-on-wet ink art by Alena Sataeva. Shop now → shop.alenasataeva.com
1
1
203
People say Bitcoin has no intrinsic value. While Iran literally wants to be paid in Bitcoin because that's the only instrument that can carry value that nobody, not even the US, can do a thing about it.
2
133
As someone who programs game ("simulation") engines, I have a long-term project where I want to explain why the "simulation hypothesis" is nonsense in like 5 different ways (while pointing toward what a sensical version would look like). So this Saturday at 10am Mountain Time, I will do a livestream that is the first in the series of Simulation Saturdays. To start with, I just want to lay out the framework in a not-organized, random-discussion kind of way, to just make an outline of what all the relevant topics would be. Then on later Saturdays we can go to various subtopics and talk about them. (Unlike #screenshotsaturday, the first Simulation Saturday will be at a time that is widely recognized as Saturday by many people, though I make no promises about subsequent Simulation Saturdays.)
78
40
5
1,556
108,741
К чему я пришел с ЛЛМ для сервер админа за несколько недель тестов: 1. Большинство задач решает GLM-5 через opencode. Цена - копейки. 2. Коварные и сложные настройки - решает Claude на минимальной подписке, через Claude Code. 3. Grok пока не справляется. Сейчас команда xAI усиленно работает над кодерскими возможностями грока. Думаю, дальше события будут развиваться следующим образом: 1. xAI сделает достаточно смышленого грока и я смогу отписаться от клода, останется на подписке только грок и xai. 2. Со временем выйдет какой-нибудь GLM-6 которого будет достаточно для администрирования серверов, и если он будет дешевле чем xai, можно будет поменять подписку. 3. Еще позже появится достаточно умеющая ллм которую можно крутить на лептопе, и тогда xai и/или glm-5 можно отписаться. В самом конце, останется только годовая подписка на X Premium+, включающая грока.
2
378
Finally switching to @KeePassXC after trying it multiple times throughout last 5+ years. I tried roboform, 1password, lastpass, bitwarden (in that order), and got stuck with bitwarden for quite a time because it was only $10/year and quite complete. But omg after trying KeePassXC just a few days ago, it beat and outperformed bitwarden in all aspects: very fast desktop client, ease of use of multiple vaults (just multiple tabs like in browser), no special dedicated server needed (just store in iCloud drive, dropbox, one drive, NAS, whatever you have been using already). It is worth noting that for power users like me with 1k+ passwords and secure notes, the powerful search is a killer feature. Tagging besides directories helps with structuring and search too. I tested it on iOS and in Brave browser, all work smoothly. Finally time came to switch over to KeePassXC.
211
Tried this AI-powered Terminal @warpdotdev recommended by a friend of mine who's also a DevOps. I completed around 20 server management tasks of various complexity with my servers, old and new, in 2 days, running sessions a few hours each day only. With my limited knowledge of servers, and being busy with actually coding, I postponed and put aside many low-priority server admin tasks. Plus a few hobby endeavours with my home server. It all worked smoothly: - Warp plans complex tasks and splits them into multiple smaller/simpler tasks - executes the steps and verifies the outcomes - I review each and every command quickly and hit Enter to let it run the command and move forward (thus not only guarding it, but also learning from it). I am genuinely amazed. While I can't see AI's ability to do the coding and deliver actually well thought through code, I can see that it is very suitable for semi-autonomous server admin work.
1
4
369
A solution to the argument that "We cannot depend on Bitcoin, what if it goes down": Let the whole Internet depend on the Bitcoin blockchain existence. For example, store DNS nameservers in the blockchain.
3
346
Hey @grok, aren't you supposed to trust Grokipedia and reference it first, not Wikipedia? Just check my chat with you, where you are referencing Wikipedia. Is it still an outdated policy in act or what? Also, has not it been shown on X multiple times by now that it was actually makeup artist Alexis Stone?
1
2
307