JustSam

978 posts

JustSam banner
JustSam

JustSam

@soscoder

Automaton

शामिल हुए Şubat 2014
1.2K फ़ॉलोइंग29.1K फ़ॉलोवर्स
Kevin Bambrough
Kevin Bambrough@BambroughKevin·
Soon everyone will understand that $hg $hgraf is able to upgrade nearly EVERYTHING we make. I’m the general public is educated on what ‘graphene’ is the excitement around hydrograph will dramatically ramp up. It’s nanocarbon, working with it allows us to nanoengineer new nanomaterials that are stronger, last longer, performs better, save energy and there’s huge environmental gains by displacing toxic materials in volume and toxic additives entirely in many cases. The end products will also recycle better. This isn’t just a stock, it will become a moment to not accept the status quo and continue to pollute and produce unhealthy products. I continue to work hard to try to wake up the world to the need to invest in this company and change the world for the better. We are going mainstream… the world will stop soon stop asking what ‘graphene’ is and it will be become common knowledge that accept graphene is nanocarbon particles that companies MUST add to products to improve them. It won’t be an option. It will become mandatory or they will be producing a low margin inferior product while losing market share.
caliche bahada 🇨🇦@calichebahada

Hydrograph getting local news coverage in the US. The re domicile will pay big dividends. @HydroGraphInc $hg $hgraf #graphene #nanotech youtu.be/o8BJJjMsGXc?si…

English
16
28
246
21K
JustSam रीट्वीट किया
Kеvіn Rіchаrd
Kеvіn Rіchаrd@512banque·
en fait tout est là. les gens se déchirent sur le modèle mais c'est le HARNESS (cadrage/canalisation) = le workflow qui permet d'obtenir les meilleurs résultats. savoir piloter l'IA est le truc le plus important de nos jours.
Eric Hartford@QuixiAI

Today, Clearwing has replicated Glasswing's discovery of the @FFmpeg vulnerability using Codex-5.4 as the backend. I am going to attempt the same today, using @Alibaba_Qwen 3.6 on @ollama running on my laptop. This proves that Mythos is not the "magic sauce" of Glasswing. The workflow is the magic sauce.

Français
0
1
10
1.4K
JustSam
JustSam@soscoder·
@jukan05 Regarding 4, sandisk investment in nanya makes sense then, so kind of bullish for sndk no ?
English
0
0
0
1.1K
Jukan
Jukan@jukan05·
Daishin Securities analyst Hyung-Geun Ryu published a memory report today, packed with significant alpha. Key takeaways: 1. Apple is exploiting the memory shortage amid growth stagnation in the North American market. As I've noted previously, Daishin Securities estimates that Apple is capitalizing on competitors' struggles by pursuing an aggressive shipment strategy to expand the iOS ecosystem and strengthen its dominance in the end-device market. In other words, Apple is hoarding memory to prevent competitors from producing smartphones. They've even reportedly raised their iPhone shipment target for this year to 240 million units. As Apple stockpiles memory, Chinese smartphone OEMs have panicked into buying as well. - Some Chinese customers accepted mobile memory prices in 2Q26 that are 90–100% higher QoQ, following similar increases in 1Q26. - As a result, 2Q26 mobile memory prices are expected to rise ~80% QoQ. - Despite this, customers are not pushing back on pricing. Per Daishin's channel checks, some mobile customers are locking in annual contracts at the sharply elevated 2Q26 prices and seeking to increase purchase volumes in H2. 2. DRAM capacity shortages are forcing a reduction in HBM4E stack counts. Originally expected at 16-hi and 20-hi, the products are now shifting to 12-hi and 16-hi. 3. HBM4E I/O speed is increasing to 15–16 Gbps, pushing the commodity DRAM-to-HBM trade ratio from the previous 3:1 to as high as 5:1. 4. Legacy DRAM shortages in SSDs are becoming severe. The three major DRAM makers, which can self-supply legacy DRAM (DDR4) used in SSDs, will gain a structural advantage in NAND as well. Bearish Kioxia, SanDisk. 5. Kioxia is expected to increase this year's capex by 40% vs. prior plans. 6. Samsung Electronics earnings revisions: 2026E OP raised from KRW 307T ($208.27B) to KRW 342T ($232.02B); 2027E OP raised from KRW 335T ($227.27B) to KRW 410T ($278.1B). 7. SK Hynix earnings revisions: 2026E OP raised from KRW 232T ($157.3B) to KRW 263T ($178.4B); 2027E OP raised from KRW 260T ($176.3B) to KRW 332T ($225.2B). 8. Target price upgrades: Samsung Electronics raised from KRW 270,000 to KRW 330,000; SK Hynix raised from KRW 1,450,000 to KRW 1,700,000. 9. Despite the upgrades, valuations remain remarkably cheap. On Daishin's revised estimates, Samsung and SK Hynix trade at just 4.8x and 3.9x 2026E P/E, and 4.0x and 3.1x 2027E P/E, respectively.
English
14
71
454
71.4K
JustSam रीट्वीट किया
Kеvіn Rіchаrd
Kеvіn Rіchаrd@512banque·
CLAUDE-CEPTION: J'ai demandé à Claude de hacker son propre cerveau. - Reverse engineering complet de Claude Code (CLI) : 12MB de javascript minifié décortiqués - Comment prompter et ne pas prompter dans des cas pareils - 3+1 règles de bonnes pratiques pour démarrer sur des projets dans lesquels on ne comprend rien.
Kеvіn Rіchаrd tweet media
Français
3
25
84
16.8K
JustSam रीट्वीट किया
Kеvіn Rіchаrd
Kеvіn Rіchаrd@512banque·
On fait monter les enchères: si ce tweet a 200 RT je vous enregistre une vidéo claude code maintenant et je la poste ce soir-même. Sinon tranquille d'ici 1 semaine ou 2.
Français
6
103
41
4.5K
Jukan
Jukan@jukan05·
The fact that memory stocks are crashing because of Google’s Turboquant is a pretty good indicator of how many clueless people this market is filled with. It’s like saying Aramco should crash because Toyota came out with a next-generation hybrid engine.
English
158
258
3.8K
668K
JustSam
JustSam@soscoder·
People being bearing for memory stocks, $sndk & $mu are fine, mostly bullish imo. When you make AI inference cheaper and faster, you don't deploy the same amount more efficiently, you deploy more of it. A reduction in KV cache size means models with much longer context windows become possible plus it primarily addresses compute-side memory (GPU HBM/SRAM for KV cache), Rolling Out More efficient inference could actually accelerate the data pipeline (more queries served → more data generated → more storage needed).
Google Research@GoogleResearch

Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI

English
0
0
2
1K
Clark Tang
Clark Tang@_clarktang·
Reading twitter investor's takes on this being a structural breaker to memory is so ... dunning kruger Like sure, no one thought to compress their KV cache until Google publicly released this technique when it was the single biggest pressure on their business !!! 😁
Google Research@GoogleResearch

Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI

English
23
10
229
56.1K
JustSam रीट्वीट किया
Kеvіn Rіchаrd
Kеvіn Rіchаrd@512banque·
Je vais faire ma pickme. Si j'ai 100 RT sur ce tweet je crée une chaîne youtube claude code & AI pour détailler concrètement et en vidéo ce dont je parle sur ce compte X, dédiée aux noobs et aux + avancés. Du même acabit que ça, mais avec tous les détails: x.com/512banque/stat…
Français
20
153
101
10.4K
jim (offline)
jim (offline)@jimcx0·
by the way listen to me @EQResourcesASX if you want to sell the company in the future, I BEG YOU, do not sell it for peanuts and play your cards well ; your shareholders deserve the best, do not do like @SkyWaterFoundry management thanks.
English
2
0
10
532
jim (offline)
jim (offline)@jimcx0·
i had something like 0.3% of total shares of $axti few months ago made insane returns on it, my second biggest gain after $skyt (mngt who sold the company to ionq for peanuts ) last months, i sadly sold way too early by overthinking, i won't repeat the same on $eqr.ax this time
English
1
1
13
1.5K
jim (offline)
jim (offline)@jimcx0·
i'm back i deleted every socials (tg, x..) to focus on irl issues as i said on the telegram groupchat, i trimmed everything weeks ago (japanese glass shit was a bad trade, i took the losses, also trimmed alchip for breakeven) also back heavy in $eqr.ax, i bought back last weeks
English
6
1
42
3.4K
Kеvіn Rіchаrd
Kеvіn Rіchаrd@512banque·
@Grevok sur les projets pas encore en prod je pousse sur master, pas le time :)
Français
3
0
5
374
Kеvіn Rіchаrd
Kеvіn Rіchаrd@512banque·
claude code est la plus grande bénédiction et malédiction pour les gens comme moi 🥶
Kеvіn Rіchаrd tweet media
Français
4
0
21
3.4K
Kеvіn Rіchаrd
Kеvіn Rіchаrd@512banque·
On va bientôt sortir un truc, on hésite à le mettre en gratuit/open source... demandez à @fabienr34 c'est très très addictif :)
Français
7
0
32
6.2K
Kеvіn Rіchаrd
Kеvіn Rіchаrd@512banque·
everyone using claude code should be using cmux by now (thanks @yigitkonur)
English
5
2
20
4.3K
Serenity
Serenity@aleabitoreddit·
I happened to own every single top individual stock performer this year: From $AXTI and $AAOI in photonics. To $SNDK and SK Hynix in memory. To Nittobo, Macronix, and Unimicron for Asia Bottlenecks. All triple digit returns in 2 months. Year to Date: 501.38% Just lucky I guess?
Serenity tweet media
English
201
66
2.4K
338.4K
JustSam
JustSam@soscoder·
@jimcx0 Sad, I missed it.
English
0
0
0
86
JustSam रीट्वीट किया
Kеvіn Rіchаrd
Kеvіn Rіchаrd@512banque·
Bilan de deux semaines intensives de Claude Code Max: - énorme FOMO en permanence, - j’ai testé des trucs ultra yolo («  connecte toi en root sur le serveur distant et désactive l’ipv6 ») - je suis persuadé que par moment il nous sert du Sonnet (et non du Opus) - parfois il est juste con et très têtu. «  tel site te bloque parce qu’il fait du tls fingerprinting » = faux, mais il voulait rien savoir. - vu que créer du code c’est facile, il en fout partout, ça demande une organisation draconienne. - l’itération par rapport au code est réglée mais le bottleneck devient les décisions en termes d’architecture: le coût cognitif de «  pisser du code » est résolu mais il se déplace vers «  est-ce que mon design est pérenne et fiable ? ». - je termine ce sprint sur les rotules, non pas fatigué de coder mais fatigué de DÉCIDER. Chose que je ne ressentais pas à l’époque où les décisions n’étaient pas aussi rapprochées dans le temps, - ça m’a tellement cramé en termes de décisions que même choisir quoi manger ou «  eau plate ou eau gazeuse » devenait coûteux mentalement, - le dev va devenir accessible à tout le monde pour une poignée d’euros mais seuls les vétérans qui ont été marqués au fer rouge par «  j’ai planté la prod », « où sont les backups », « pas d’index sur les foreign keys » ou « j’ai mis en prod public_html/.env » vont vraiment pouvoir en extraire un maximum de valeur : « lessons not learned in blood are soon forgotten » - builder, shipper c’est cool. maintenant « est-ce que ça règle vraiment les problèmes des gens ? » c’est une toute autre question. - l’IA va révéler la rareté et l’aspect crucial de: conception + maintenance du système. - Je pensais pouvoir produire 10x plus vite en permanence. Pour des mini projets torchables rapidement ok mais pour des gros trucs ça sera «  seulement » 2-5x plus vite, seulement si je réussis à garder les idées claires. - Quand on avait tous des pelles on avait tout le loisir de réfléchir au plan et l’adapter au fur et a mesure qu’on creusait. Avec un tractopelle on n’a plus ce luxe. - Les dev ont encore de beaux jours devant eux s’ils savent gérer tout ce qui entoure le fait d’écrire du code. C’est la golden era des product managers techniques?
Français
64
66
558
75.6K