maxgreco (@maxgreco) header
maxgreco (@maxgreco) profile picture

sysadmin, web manager, photographer, gamer, bouncing between my tech and humanistic side, Biella - Italy

Biella - Italia
Joined April 2008
Photo by maxgreco (@maxgreco)Photo by maxgreco (@maxgreco)Photo by maxgreco (@maxgreco)Photo by maxgreco (@maxgreco)Photo by maxgreco (@maxgreco)Photo by maxgreco (@maxgreco)Photo by maxgreco (@maxgreco)Photo by maxgreco (@maxgreco)Photo by maxgreco (@maxgreco)Photo by maxgreco (@maxgreco)
maxgreco retweeted
🦔ING tried to sell a $500 million bundle of data center loans this week and couldn't find buyers at a decent price. Now the Dutch bank is trying to sell them one at a time. My Take Banks lend the money to build these data centers, then sell the loans to other investors so they can lend again. That only works if buyers show up, and they're getting full. HSBC already dumped its piece of the Stargate site in New Mexico, and Barclays bought insurance on its AI loans. Banks don't do any of that when they feel good about a loan. Big Tech still plans to borrow hundreds of billions more next year, and I think the weakest projects lose their funding first, which means fewer chip orders for Nvidia in 2027. Hedgie🤗
37
117
11
478
16,495
maxgreco retweeted
Project Maya v1.1.73 is out with more hardware support to Inkling 2 Mini 😊 ✅ Inkling 2 Mini now runs on NVIDIA cards from 8 GB: the experts that don't fit stay in RAM ✅ Inkling 2 Mini also runs on V100 and RTX 20 cards, and on AMD on Linux (Radeon and Strix / Gorgon Halo) ✅ Drafting for Inkling 2 Mini only where it is faster ✅ Fixes: the VRAM reserve is kept, and Hy4-Flash runs on more two-GPU PCs Full notes 👇 github.com/mw00/project-maya…
Project Maya v1.1.72 is out: @thinkymachines Inkling 2 Mini joins the family! This is the result of the community getting together and @superalesha taking the initiative, now we all can enjoy a new model running faaaast. 🚀 ✅ Inkling 2 Mini arrives as an experimental preview on its own engine: the whole model on one 24 GB NVIDIA card (Linux) ✅ Measured on one RTX 3090: 124.6 tok/s decoding, 2,542 tok/s prompt reading ✅ Thinking levels, tool calls and pictures, through the OpenAI and Anthropic APIs github.com/mw00/project-maya…
7
7
2
73
3,306
今天才知道一个颠覆认知的硬核事实: 俄罗斯人翻墙,竟然压根不用 VPN!!🤯 在俄罗斯技术圈,大家最推崇的根本不是传统代理,而是一款名叫「Zapret」的纯本地开源翻墙工具。 它的路子野到什么程度? 既不需要花钱买海外 VPS 也不用折腾任何机场订阅 更不依赖境外代理服务器 直接装在 Windows、Linux 或者 OpenWrt 软路由上就能本地开跑 核心原理说白了就是「本地硬刚」: 在网卡发包的那一瞬间,直接切分并修改网络数据包特征,精准干扰运营商的 DPI(深度包检测)设备,让审查系统看花眼,从而直接绕过部分网络封锁和限速。 对比一下两边的画风简直太真实了: 中国人翻墙,天天钻研 Reality、HY2 协议怎么伪装得滴水不漏,到处抢 VPS、测机场线路 俄罗斯人翻墙,连中转服务器都懒得搭,直接研究怎么在本地把审查设备给绕晕! 战斗民族果然是能正面解决问题,绝不跟你玩躲猫猫!!!🤣🔥 github.com/bol-van/zapret
103
172
12
1,523
160,656
Sam Altman realizing the AI bubble has burst before he could take OpenAI public
28
56
9
984
31,772
maxgreco retweeted
The super powers of AI is putting the spirited individual in the center. The Idea Guy suddenly has an avenue to push their dreams that doesn't rely on raising big money or being a master programmer. This is very exciting.
Ruby on Rails creator DHH says AI is bringing back the individual hacker's ability to shape the future of computing: "I love that we have this energy around computers again, like it was in the '90s, the '80s, where hackers and experimenters are suddenly at the forefront of everything." "I love this energy that now individuals, thanks to the power of AI, can actually play a pivotal role in the forward progression of the industry." "Once the internet got sort of bottled up and commercialized, a lot of the frontier work to push things forward just got gobbled up by the big companies." "We closed off all these domains of the consequential bits of the computer. And only the biggest companies in the world could work on them. And now we're re-democratizing it. We're returning it to the hacker culture. We're returning it to the individual who can actually move some of this stuff forward." @dhh
177
169
37
2,249
79,089
Some of you thought that Inkling 2 Small was too big at 594B parameters. We heard you: Inkling 2 Mini is a looped transformer with just 26B parameters, small enough to run on your laptop, powerful enough for agentic work with a score of 20.1 on AA Index. thinkingmachines.ai/news/ink…
640
1,698
1,037
18,384
3,293,508
maxgreco retweeted
Inkling 2 Mini from Thinking Machines! 🔥 - 26B total params, 3.4B active (5.4B looped) - Weights just 20 GB - Small enough to run on a laptop - Ahead in 10 of 13 in coding vs Qwen3.8-27B. I'll try it out on my dgx spark. Link below.
66
68
19
1,122
55,458
maxgreco retweeted
Here's my updated model tier list Google is back big time
127
30
12
1,141
74,434
Per #Tajani sulla morte suicida di Gianni #DeGennaro i processi si fanno solo in tribunale, ma un mese fa, invece, bastava la richiesta di custodia in carcere per proclamare la condanna di #YoussefSiher. #GianniDeGennaro
1
21
71
636
maxgreco retweeted
I'm hearing from my friends at OpenAI that the vibes ARE DOWN. Like almost depressed. This might be the worst week in the company's history type bad. Tuesday was supposed to be their big day. They launched GPT-6.2 and at the same time dozens of early testers dropped the worst reviews imaginable. Theo's review got more views than the launch post. This went even worse than the GPT-5 launch. Then Google chose violence and dropped the opposite of a GPT model with Neon. It's conversational, understands intent, very good at coding/3D and brought them back to the top. A lot of people say they already passed OpenAI. And the worst thing: Nano 🍌 Pro 2 is currently in the arena and the outputs are stunning. So Google basically overtook OpenAI in a single week. For months OpenAI tried to pretend to be the best lab by bullying mathematicians and burning massive amounts of compute on Navier-Stokes (and countless other math problems). Now Anthropic mogged them by solving the Riemann Hypothesis and makes OpenAI look like an undergrad while they are doing 250k-page alien math. The last funding round was a disaster too. They wanted $30B at $1.5T. They got $15B at $1.25T. They lost $10.9B last quarter, so this money only lasts them ~4 months. They are basically cooked and the valuation will likely drop further. OpenAI isn't the best at anything anymore. They might genuinely be done.
458
486
195
8,971
1,281,865
maxgreco retweeted
New NVIDIA paper finds that once an LLM has reasoned through a problem, the solution can run as CPU code for rephrased and re-numbered versions of it. that answers similar questions 16x to 291x cheaper. LLMs re-solve every problem from scratch, even when only wording or numbers change. Agents sending the same instructions with new data pay for that repeated reasoning each time. Their method has Claude Opus 4.5 turn a reasoning trace into small Python functions, with changeable inputs as parameters. At run time they execute on a CPU, calling a fine-tuned 1B model only for small text jobs like extracting values. The cache matched or beat the source model's accuracy on 7 of 8 math and science benchmarks, scoring 93.3% versus 83.3% on AIME '25. It handled at least 90% of rephrased or re-numbered variants. If your agent runs the same prompt on new inputs all day, cache the reasoning as code and use the LLM only on misses.
65
155
34
1,330
50,972
maxgreco retweeted
In Italia si tassa in maniera indecente il lavoro mentre le rendite e i grandi patrimoni vengono tassati poco o nulla. È uno squilibrio che punisce chi lavora creando ricchezza e premia i più ricchi che estraggono ricchezza senza produrla. Un ostacolo alla crescita economica.
12
53
731
maxgreco retweeted
Thinking Machines is the one of the few American organisations with a moral backbone It is our duty to help protect and support them and spread their work! Inkling mini matches Opus 5.5 on Cyber bench
Replying to @0xSero
Inkling 2 mini is scoring the same as opus 5.5 on cyberbench!?!?
34
97
5
2,264
35,715
maxgreco retweeted
I realized I hadn't done the 100 HTML experiment for Qwen3.8-Flash-Next yet, so here it is. I asked Qwen3.8-Flash-Next to create 100 HTML files. Rules were simple: - look stunning - zero repetitive designs - go full creative mode I said before that this model shines in frontend, and it really does. For a model that runs comfortably on a 128 GB machine like a DGX Spark or a Mac, it does exceptionally well! You could probably build ANYTHING with this model. It's THAT good. For web design specifically, I think it's one of the better open models out there. Just look at the results yourself: miaai-lab.github.io/Qwen3.8-… See how other models performed: mia-ai.net/lab/experiments/1… Some files I personally liked: Rain on Glass miaai-lab.github.io/Qwen3.8-… Brass Orrery miaai-lab.github.io/Qwen3.8-… Ants miaai-lab.github.io/Qwen3.8-… Geoboard miaai-lab.github.io/Qwen3.8-… The Sealed Terrarium miaai-lab.github.io/Qwen3.8-… The Pentatonic Parlour miaai-lab.github.io/Qwen3.8-… The MARVEL of the Stairwell miaai-lab.github.io/Qwen3.8-… Baitball Frenzy miaai-lab.github.io/Qwen3.8-… Fire Escape at Dusk miaai-lab.github.io/Qwen3.8-…
41
14
3
314
31,131
maxgreco retweeted
Ladri che piangono quando vengono derubati
2
12
289
La presidente del Messico Claudia Sheinbaum Pardo è andata e tornata da Shenzhen, Cina, 21 ore di volo, per il vertice APEC 2026, in classe economica. Nel 2023, il Messico ha venduto il Boeing 787 presidenziale, utilizzando il ricavato per la costruzione di due ospedali pubblici.
25
145
4
608
10,053
My Journey to Build an Affordable AI Machine - Part III The shopping list is coming together: ✅ Step 1 RAM: 128GB DDR4-3600 £711 ✅ Step 2 PSU: Corsair HX1500i (1500W) £201 ✅ Step 3 CPU: AMD Threadripper 3960X £410 ✅ Step 4 Motherboard: Gigabyte TRX40 DESIGNARE £390 ⏳ Next - GPU Finding the right motherboard took far longer than I expected! If I only wanted to run one GPU, the choice would have been easy. But my goal is to support 2× RTX 3090s with NVLink, which makes PCIe bandwidth, slot spacing, cooling and motherboard layout much more important. After plenty of research, I finally found the CPU and motherboard on eBay UK at reasonable prices. 💡 Along the way, I compared five platforms: X99 | B550 | X570 | TRX40 | WRX80 I've put together a comparison table covering PCIe bandwidth, RAM options, power requirements and approximate UK costs. X99 is the cheapest and plenty of DDR3 supplies. I guess £1200 if coupled with 3060 or 5060. Sharing it here for anyone planning an 'affordable' local AI workstation. Hopefully, it'll save you some of the research time I spent!
My journey to build an affordable AI machine. Step 1. Bought 128gb DDR4-3600 checked ✅ £711 Step 2. Power supply HX1500i ✅ £201.99 [Prepare for 2xGPU] Next: Motherboard [TRX40/X570/X99] ? I am considering TRX40 as it give me fast PCIe speed and it allow me to copy next-flash expert twice as fast Note: All of these can be half price like 64gb (or 3200MHz DDR4 ram), and 750w power supply. I just want to build once for all. Anyone bought motherboard from ebay China or Aliexpress?
19
4
66
5,667
maxgreco retweeted
This prediction of mine wasn't accurate and for that I apologize Inkling 2 Mini gives you Opus 4.5 intelligence and capabilities on an RTX 3080 (~300$ USD a year ago) Local AI is winning BIG TIME
Prediction We will have Claude Code + Opus 4.5 quality (not nerfed) models running locally at home on a single RTX PRO 6000 before the end of the year
82
131
7
2,879
95,008
maxgreco retweeted
My magnum opus, my greatest production. This skill was forged through thousands of hours of inference optimisation I have collected 30 posts of people who used this to get anything from 30% speed up to 5x speed ups Prove it works, you will get a shoutout github.com/sybil-solutions/i…
65
102
6
1,862
52,567