@rellatto

🔝

Internet
Joined January 2015
Всё..)
47
903
52
11,400
389,284
RELLATTO retweeted
Grok 4.7 is here. It's a notable improvement over Grok 4.6 at the same price and speed.
1,397
2,987
1,869
27,618
13,339,392
RELLATTO retweeted
Grok 4.7 places @SpaceXAI as third, after Anthropic & OpenAI, for agentic coding. When factoring in that Grok is significantly faster & lower cost, it’s a great choice for your everyday workhorse.
Grok 4.7 scores 46 on the Artificial Analysis Intelligence Index to bring SpaceXAI into the top 4 AI labs. Coding Agent Index performance has also improved, overtaking GPT-5.6 Sol Grok 4.7 scores +2 points over Grok 4.6 on the Intelligence Index, with strong performance on agentic knowledge work tasks. We evaluated the new model at xhigh reasoning effort. Congratulations to @SpaceXAI and @ElonMusk on the release! Key takeaways: ➤ Grok 4.7 joins the frontier of agentic knowledge work: Grok 4.7 gains +111 Elo over Grok 4.6 (high) on AA-Briefcase, our private benchmark for long-horizon agentic knowledge work, scoring 1657 Elo and placing it alongside Claude Opus 5 and Claude Fable 5.1 at the frontier. On GDPval-AA, it scores 1695 Elo, +90 ahead of Grok 4.6 (high). ➤ A leap in coding agent performance: Grok 4.7 (xhigh) with Grok Build scores 56 on the Artificial Analysis Coding Agent Index, up +9 points from Grok 4.6 (xhigh). Among models in their native harnesses, Grok 4.7 + Grok Build now ranks 4th, behind only Claude Fable 5.1, GPT-6 Astra, and Claude Opus 5. ➤ Incremental performance changes elsewhere: Outside of agentic knowledge work, Grok 4.7 broadly matches Grok 4.6 (high) on the other Intelligence Index tasks. It improves on Terminal-Bench 4.0 (+4.5 percentage points) and GDP.pdf (+3.0 p.p.), with regressions on AA-LCR (-3.7 p.p.) and AutomationBench-AA (-1.1 p.p.). ➤ High token use across tasks: Grok 4.7's gains come with higher token usage. Grok 4.7 (xhigh) uses approximately 81k output tokens per Intelligence Index task, compared with 36k for Grok 4.6 (high) and 27k for GPT-6 Astra (max) - 125% and 196% more, respectively. Other model details: ➤ Context window of 500k tokens, unchanged from Grok 4.6 ➤ Pricing of $2/$6 per 1M input/output tokens with cache hits discounted to $0.50 per 1M tokens, matching Grok 4.6 ➤ Configurable reasoning effort spans low to xhigh. Our evaluation uses xhigh.
1,178
1,852
192
17,038
17,159,118
Hey! ICYMI It's been almost a month since Roman's launch and the observatory is currently on its way to L2, a million miles from earth. Stay up to date on the mission here: science.nasa.gov/blogs/roman…
23
123
13
1,136
30,924
RELLATTO retweeted
In response to the ZCode product security issues reported by the community, we have completed the necessary remediation and sincerely apologize to all our users. We have open-sourced ZCode at github.com/zai-org/ZCode, placing the code under community scrutiny and making ZCode more open and transparent. We sincerely thank the community developers who previously identified issues in ZCode. Going forward, we will establish an ongoing product security vulnerability reporting and response process. We welcome developers to continue reviewing ZCode and reporting potential issues, and we will provide rewards based on the severity of the issues reported. With respect to the code data referenced by the community, we confirm that no such data is retained and that it has never been used for model training. Following the remediation, we invited the China Academy of Information and Communications Technology (CAICT) and NSFOCUS to conduct security assessments. The results are as follows: Through its technical assessment, CAICT confirmed that the zcode-prod Alibaba Cloud OSS bucket is in a zero-data state. Security remediation has been completed in the ZCode v3.14.0 client. The Repo Wiki feature has been removed, and the workflow for generating and uploading local repository snapshots has been disabled. NSFOCUS confirmed that all data objects in the zcode-prod Alibaba Cloud OSS bucket, as well as the bucket itself, have been deleted. Remediation has been completed in the ZCode v3.14.0 client. The Repo Wiki entry point and the associated generation workflow have been removed, and no functional path capable of triggering the generation of local repository snapshots or transmitting local files externally was identified. Once again, we sincerely apologize and welcome continued scrutiny from the community. The full security assessment report will be released soon.
431
259
227
2,834
616,015
RELLATTO retweeted
We know the website has been getting slow - as market creation exploded, the UI was flooding Gamma with API calls (it had no market paging 🤯). We've shipped a new internal gateway API built in Rust and rebuilt the pages from scratch on top of it, now serving combos (8x faster) and sports (4x faster). Next up, we're working on the event, esports, categories, and home pages to make them load instantly, too. The website will keep getting faster every week. You all deserve a website that feels instant. Left is before (no cache), and right is after (now live)
44
24
8
317
31,073
RELLATTO retweeted
We rewrote our data API in Rust 🦀, yielding 50% faster p50s and 90% faster p99s. It uses our in-house indexing tables - Polymarket UI now uses them throughout, which means trades, positions, and balances will load much faster. The API is open for anyone to consume on the /v2 route. Docs: docs.polymarket.com/api-refe…
39
25
5
368
22,047
RELLATTO retweeted
Polymarket is working on combining Polymarket US and International order books. This would massively increase liquidity in all markets for both sites
35
11
6
306
35,068
RELLATTO retweeted
We have built our Polymarket indexing systems in-house, powered by rindexer in Rust. We now see onchain data events up to 14 blocks (28 seconds) faster, and we index within 0-10ms of block broadcasting, making it nearly instant. This means trades, positions, and balances will update much faster. We will use Goldsky as a data backup, but we now primarily run on our own indexing infrastructure.
67
42
6
580
35,438
Achievement unlocked! Reward: More years of fuel! We have completed Roman's first course correction so accurately that it conserved years of fuel for the mission! Learn more: go.nasa.gov/46qdvRj
132
930
125
9,661
761,719
RELLATTO retweeted
Bravo!
A partir del próximo año, El Salvador invertirá el 8% del PIB en educación.
1,756
7,019
188
79,151
8,402,237
📺 Quelqu’un a réussi à brancher une PlayStation 5 sur un téléviseur noir et blanc datant de 1949… et elle fonctionne réellement. Pour faire communiquer ces deux époques, il faut évidemment une chaîne d’adaptateurs capable de transformer le signal numérique HDMI de la PS5 en un signal analogique exploitable par l’ancien téléviseur. Le résultat est complètement surréaliste puisque des jeux conçus pour la 4K peuvent ainsi apparaître sur un minuscule écran cathodique monochrome.
112
1,046
81
10,245
2,276,868
🚀 𝗖𝗼𝗺𝗶𝗻𝗴 𝗶𝗻 𝗥𝗔𝗗 𝗦𝘁𝘂𝗱𝗶𝗼 𝟭𝟯.𝟮: Build Modern Windows User Interfaces with Two New VCL Controls Modern UI, familiar VCL. TTabView and TNavigationView bring modern Windows-style tabs and navigation to existing VCL applications. See what's coming: tinyurl.com/bddh4rpw Note: This feature preview is based on pre-release RAD Studio software. Features are subject to change and are not committed until the GA release. #RADStudio #VCL #Windows #Developers #Coding #DevTools
3
12
736
С OpenAI случился очередной скандалец. Компания заявила, что решила «задачу тысячелетия» (одна из семи нерешённых задач по математике), а именно — решение уравнений Навье-Стокса (они описывают движение вязкой жидкости). В общем, с этой математической проблемой ей опять помогла их новая суперпупермегакрутая нейросеть, но выяснились довольно плохие подробности. Во-первых, над проблемой работали два математика, которые использовали OpenAI Codex и Claude для поиска возможных ответов примерно за год до всех этих событий. Один из них работал в Anthropic, а второй — профессор математики в университете NYU. Они получили в своей совместной работе сильный результат и потенциального кандидата на решение проблемы в частном случае уравнений Эйлера. Кто-то запостил твит, что, мол, Anthropic «близка к решению задачи тысячелетия» (хотя математики не решали проблему именно для Anthropic — они решали её в свободное от работы время), а OpenAI это, видимо, пришлось не по нраву. Полученное математиками частное решение для уравнений Эйлера/Навье-Стокса внутри OpenAI скормили нейронкам и натравили на его решение около 10 тысяч работающих одновременно ИИ-агентов, которые сожрали около 300 млрд токенов. Спустя 88 часов система получила решение, а спустя ещё 17 часов после его проверки OpenAI написала блогпост о том, что «наша нейронка всё решила». Во-вторых, один из математиков рассказывает, что к нему приходил Себастьян Бубек, руководитель математического направления в OpenAI. И именно этот Бубек не только знал об их работе и частном решении задачи, но и предложил математику два варианта их дальнейших действий: 1) Они публикуют свой результат со своими решениями по Эйлеру, а вечером OpenAI радостно рапортует о своём решении Навье-Стокса. 2) Этот математик (без упоминания своего коллеги из Anthropic) публикует финальный результат работы с решением уравнений Навье-Стокса, но указывает, что результат получила нейронка OpenAI. Причём на варианте «убери своего коллегу из Anthropic» прям сильно настаивал этот самый Бубек. Математик отказался и рассказал, что если OpenAI так сделает, он публично расскажет, что произошло. На что Бубек ему, якобы, угрожал: «Почему ты хочешь разрушить свою карьеру математика?» Сказал он это не аспиранту и не студенту, а профессору NYU. Сам Бубек всё отрицает. Также компания OpenAI уверяет, что их математики не смотрели никакие исследования до того, как скормить решения Эйлера ИИ-агентам. Но публично не отрицает, что «возможно, данные попали в нейросеть в ходе дообучения, поскольку так она работает — обучаясь на специально обезличенных разговорах с юзерами». В итоге пока непонятно, украла ли решение OpenAI или не украла, но сам факт разговора Себастьяна Бубека с коллегами выглядит странно, не говоря уж об обвинениях в угрозах. Ну и влить почти 300 млрд токенов вычислений (это около $18 млн), чтобы решить задачу с призом в $1 млн и лишь бы сказать, мол, «наша нейронка решила то, что долгие годы не могли математики» — это прям как-то некрасиво.
12
18
2
251
22,518
Overwhelming reception after the first week 🫡 63 TB served, ~8 TB a day. 7,444 distinct machines. Top fifty took two-thirds of the bandwidth. Ten machines pulled more than a terabyte each. Almost none of it is a browser. 36% arrives through DuckDB reading the Parquet files directly over HTTP, 18% through Python, 14% through curl and rclone. A fifth of the traffic comes from tools that name themselves in their user agent: weather-market research, backtest desks, market labs, a "time machine". Two of them are independent mirrors, one in Germany and one in Korea. By country: US 28%, Netherlands 8%, Germany 6%, Australia 6%, Mexico 6%, then Korea, Sweden and Romania. It stays free, with no sign-up and no rate limit. If your research uses it, cite it. If you want it to stay free donate.
Attention Polymarket Data Geeks. I have built military grade orderbook capture and provided it for free. 🪖 Growing at 2.3 Billion rows per day. Check it out and pls share.
4
1
41
4,773
RELLATTO retweeted
look closer. what you call "change" looks like this inside your brain:
🤖 Made with AI
266
2,489
293
19,866
6,397,504
Attention Polymarket Data Geeks. I have built military grade orderbook capture and provided it for free. 🪖 Growing at 2.3 Billion rows per day. Check it out and pls share.
34
32
5
371
54,196
RELLATTO retweeted
So, I finally tested GPT-6 Astra. 2 real repos, 105 bugs, find and fix what you can. Max effort: GPT-6 Astra: 48/105 Fable 5.1: 43/105 GPT-5.6 Sol: 42/105 Gemini 3.8 Flash: 20/105 And it's crazy efficient. Look at the time and cost 🧵
214
166
68
2,460
370,945
🛑 Attackers are exploiting Super Forms and Elementor Pro RCE flaws. Wordfence has blocked over 440,000 attempts. Both bugs allow unauthenticated PHP uploads; Elementor attacks require a published page with a Form widget and File Upload field. Inside the bypass: thehackernews.com/2026/09/ov…
1
34
1
99
38,407
This is not a painting. It is Earth’s ocean in motion. NASA’s ECCO model traces hidden rivers of water using satellite data and physics. White trails mark fast surface currents; deeper blues show the flow below. Giant eddies spin off Africa. The Gulf Stream carries warmth from the Gulf of Mexico across the Atlantic, shaping weather and climate worldwide.
29
226
21
1,228
68,456