Ofer Shaal

I got so obsessed with RuVector (by @rUv) that I built an entire game around it. 🏎️ Vector Vroom, AI race cars + brain + full tutorial, right in your browser. Watch the cars learn and improve in real time. Open source. Dangerously addictive. PRs welcome! vectorvroom.shaal.dev

@NZXT Pick me

Who wants this message from us?

@NVIDIAGeForceUK #007FirstLightRTX 🤞

Recruits, your first prize is here... A custom GeForce RTX 5080 Founders Edition + PC copy of the game. Comment #007FirstLightRTX to win 👇

@NVIDIAGeForce #007FirstLightRTX 🤞

Recruits, your first prize is here... A custom GeForce RTX 5080 Founders Edition + PC copy of the game. Comment #007FirstLightRTX to win 👇

A man with no working truck convinced Wall Street he had built the next Tesla. His company hit $30 BILLION. All he did was push it down a hill with no engine. > Trevor Milton founded Nikola in 2014, named after the same inventor as Tesla. > The goal was to build hydrogen powered trucks that would make diesel obsolete. He had no trucks. > In 2018 he released a promotional video called Nikola One In Motion. It showed a sleek semi truck accelerating smoothly down an open highway. Investors went wild. > What nobody knew was that the truck had no engine, no fuel cell, and no propulsion system of any kind. > Milton's team towed it to the top of a hill, tilted the camera to hide the slope, and let it roll. > He spent the next four years doing the same thing with words. On podcasts, television and social media. > Investors were told Nikola could produce its own hydrogen. It could not. They were told the trucks were ready for production. They were not. They were told orders were flooding in. They weren't. > In June 2020 Nikola went public. Within days the company was worth $30 BILLION, more than Ford. > Milton's personal stake hit $7.3 BILLION overnight. > A $32.5 MILLION ranch in Utah followed. A record for the state at the time. > In September 2020 Hindenburg Research published a report calling Nikola "an intricate fraud" built on "an ocean of lies." Milton resigned within ten days. > A federal jury convicted him of securities fraud and wire fraud in 2022. Sentenced to four years in prison the following year. > He never went. He was free on $100 MILLION bail pending appeal. > He and his wife donated $3.2 MILLION to Donald Trump's 2024 campaign. > In March 2025 Trump gave him a full pardon. The pardon erased $168 MILLION in restitution to defrauded shareholders. > Nikola filed for bankruptcy the following month, leaving thousands of investors with nothing. The company never had a product. The only thing that was real was the $30 BILLION valuation, the $7 BILLION that landed in his pocket and the pardon that made sure none of it had to be returned.

@nir_benz x.com/AIatMeta/statu…

משפחת המודלים Llama-4 של Meta שוחררה לפני שנה. כמה מחשבות וקצת היסטוריה על אבן הדומינו שנפלה ברגע הזה: - פרט DeepSeek R1 שיצא חודשיים לפני כן, המודלים של מטא היו השליטים של LLMים בקוד פתוח. Qwen, MiniMax, Kimi, GLM - העלייה שלהם קרתה רק אחר כך (נכון ללפני שנה המודלים הסינים פרט לדיפסיק היו לא רלוונטים). - זה היה ריליס סופר מרגש. כולם חיכו לו. ללאמא-3 היו משפחת מודלים שכבר ראתה אימוץ בהרבה מוצרים. - והמספרים בהתחלה היו מטורפים. - אבל אז הסתבר שהם שילוב של פיברוק, benchmark hacking והדובדבן שבקצפת - דיסטילציה מעל אותו דיפסיק כדי לעמוד ביעדים. - ואז החל שיטסטורם. חוקרים שעבדו על המודל התחילו להכחיש את הקשר שלהם אליו. - צוקרברג לא אהב את זה. כאן החלה השקיעה של ארגון המחקר הוותיק של מטא, מהלך שהתגלגל בסופו של דבר לרכישה של Scale ב-14 מיליארד דולר, הכניסה של Alexandr Wang לחברה יחד עם עוד עשרות חוקרים במענקים של עשרות מליונים לחוקר - ובסוף גם העזיבה של יאן לקון, יחד עם חלק ניכר ממה שהיה בעבר FAIR. - זה אותו FAIR ששלט במשך עשור בקוד פתוח, בפרט בראייה ממוחשבת אבל גם בשפה. מכון מחקר שהביא יכולות על אנושיות בפוקר, את Segment Anything Model וכמובן RetinaNet לזקנים בנינו. - גוגל, שלפני שנה לא הייתה בכלל במשחק, מחזיקה היום את המודלי קוד פתוח הכי טובים בשוק. - ולגבי מטא? מתחדד הרושם שהם שרפו מיליארדים על כלום ושומדבר. פאקינג שנה.

@heynavtoor well, @grok 4.2 fast solved it correctly - grok.com/share/bGVnYWN5…

🚨SHOCKING: Apple just proved that AI models cannot do math. Not advanced math. Grade school math. The kind a 10-year-old solves. And the way they proved it is devastating. Apple researchers took the most popular math benchmark in AI — GSM8K, a set of grade-school math problems — and made one change. They swapped the numbers. Same problem. Same logic. Same steps. Different numbers. Every model's performance dropped. Every single one. 25 state-of-the-art models tested. But that wasn't the real experiment. The real experiment broke everything. They added one sentence to a math problem. One sentence that is completely irrelevant to the answer. It has nothing to do with the math. A human would read it and ignore it instantly. Here's the actual example from the paper: "Oliver picks 44 kiwis on Friday. Then he picks 58 kiwis on Saturday. On Sunday, he picks double the number of kiwis he did on Friday, but five of them were a bit smaller than average. How many kiwis does Oliver have?" The correct answer is 190. The size of the kiwis has nothing to do with the count. A 10-year-old would ignore "five of them were a bit smaller" because it's obviously irrelevant. It doesn't change how many kiwis there are. But o1-mini, OpenAI's reasoning model, subtracted 5. It got 185. Llama did the same thing. Subtracted 5. Got 185. They didn't reason through the problem. They saw the number 5, saw a sentence that sounded like it mattered, and blindly turned it into a subtraction. The models do not understand what subtraction means. They see a pattern that looks like subtraction and apply it. That is all. Apple tested this across all models. They call the dataset "GSM-NoOp" — as in, the added clause is a no-operation. It does nothing. It changes nothing. The results are catastrophic. Phi-3-mini dropped over 65%. More than half of its "math ability" vanished from one irrelevant sentence. GPT-4o dropped from 94.9% to 63.1%. o1-mini dropped from 94.5% to 66.0%. o1-preview, OpenAI's most advanced reasoning model at the time, dropped from 92.7% to 77.4%. Even giving the models 8 examples of the exact same question beforehand, with the correct solution shown each time, barely helped. The models still fell for the irrelevant clause. This means it's not a prompting problem. It's not a context problem. It's structural. The Apple researchers also found that models convert words into math operations without understanding what those words mean. They see the word "discount" and multiply. They see a number near the word "smaller" and subtract. Regardless of whether it makes any sense. The paper's exact words: "current LLMs are not capable of genuine logical reasoning; instead, they attempt to replicate the reasoning steps observed in their training data." And: "LLMs likely perform a form of probabilistic pattern-matching and searching to find closest seen data during training without proper understanding of concepts." They also tested what happens when you increase the number of steps in a problem. Performance didn't just decrease. The rate of decrease accelerated. Adding two extra clauses to a problem dropped Gemma2-9b from 84.4% to 41.8%. Phi-3.5-mini from 87.6% to 44.8%. The more thinking required, the more the models collapse. A real reasoner would slow down and work through it. These models don't slow down. They pattern-match. And when the pattern becomes complex enough, they crash. This paper was published at ICLR 2025, one of the most prestigious AI conferences in the world. You are using AI to help you make financial decisions. To check legal documents. To solve problems at work. To help your children with homework. And Apple just proved that the AI is not thinking about any of it. It is pattern matching. And the moment something unexpected shows up in your question, it breaks. It does not tell you it broke. It just quietly gives you the wrong answer with full confidence.

@bios_hazard Can I have that mode for a human please?

so I've got a bajillion apps on my android phone. I learned today I can use focus mode to pause almost all of them. So I just have focus mode on now and wow it's so fast now!!

With all the recent AI advances, how many XKCD comics have become obsolete and how many are still just as true as ever? xkcd.com/1205/

brute-code > vibe-code

@GeoffreyHuntley Do you suggest building it using Claude Agent SDK? Or building it from scratch?

building an coding agent remains one of the things that delivers the most alpha skip the advent of code clone cursor instead you’ll have something barebones in under 15 mins and from that point you are recursively building an agent with an agent signed, i’m building another code agent

@OfficialLoganK @GoogleAIStudio I'd love to try it

Big upgrade to vibe coding in @GoogleAIStudio lands in Jan, but if you want to test early… 👇🏻

@Pokee_AI Pokee

🚨 We’re reaching out to a few of you in DMs with early access to something huge. This is one of the first major open-source drops of its kind in the U.S., and it’s almost here. If you’ve already heard from us, you’re in. If not, no worries! we still have a few spots left before launch next week. 👇 Drop “POKEE” in the comments to lock in early access before we go live!

@_adishj Mosaic

today, we're launching Mosaic, the agentic video editor. in a world tending towards AI slop, create something real. no waitlist — public beta is now live at mosaic [dot] so. comment "MOSAIC" to get 1,000 free credits dropped into your account. this release comes with 7 key features (thread):

Introducing VEED Fabric 1.0. The world’s first AI talking video model. Built for creating great-looking videos, FAST. • Create 1-minute-long videos • 60x cheaper • 7x faster Repost = Full guide in DMs.

@viktor_vrp cc:@connorkapoor x.com/connor_kapoor/…

Anyone know a fast CNC milling service? 1-2 days turnaround.

@Sumanth_077 @threadreaderapp unroll

You can now make Claude Code, Gemini CLI, and other coding agents 10x more powerful by giving them long‑term memory! It just takes a single line of code. Here’s a step‑by‑step breakdown (100% local):

just shipped taskmaster v0.18 🚀 → @claude_code provider without API keys → init with IDE-specific profiles for @cursor_ai @claude_code @windsurf_ai @roocode @cline @code @Trae_ai → cleaner git repos & Python support → PROVIDER_BASE_URL follow + bookmark + dive in 👀👇

just shipped taskmaster v0.17 🚀 → multi-context task management via tagged lists → `research` MCP tool with interactive CLI → get-task/show upgrades & batch ops → --append for update-task → save research to docs → multi-status filters follow + bookmark + dive in 👀👇

I'm on the waitlist for @perplexity_ai's new agentic browser, Comet: perplexity.ai/comet