Algorithm (@Pro_Algorithm) header
Algorithm (@Pro_Algorithm) profile picture

Breaking AI & tech news, game-changing launches and free open-source GitHub tools you need to see first. Stay ahead of the algorithm.

Joined November 2026
Photo by Algorithm (@Pro_Algorithm)Photo by Algorithm (@Pro_Algorithm)
ONE HOUR. ONE CREATOR. ONE AI. A WHOLE ORIGINAL GAME. This is the part of the AI wave nobody is ready for. Creator Rubzem says he built this original physics platformer with AI in just 1 hour. The twist? You play as SHAPESHIFTING WATER, solving puzzles as you go. And it gets crazier: - Entirely procedural - The whole game is only 190 KB - Made with Fable 5.2 - Playable right in your browser 190 KB. Not MB. KB. Solo game dev is being rewritten in real time, and the gap between "idea" and "playable" just collapsed to an afternoon. Would you play this?
4
5 YEARS OF HUMAN WORK vs 5 MINUTES OF AI. Luke Hutchison - former Google AI research scientist, MIT/Harvard PhD - says he just built what he calls the world's fastest SAT solver with Google's Gemini 4 Neon. His story: - 15 years ago he spent five years and thousands of hours building a SAT solver. It worked, but it was slow vs the state of the art. - He handed Neon all his old code. It analyzed it, kept his best ideas, and built a hybrid solver in LESS THAN FIVE MINUTES. - A few more prompts: 50x faster. Then 10x faster again. Then 100x faster again. He named it Frobenius and claims it's on average thousands of times faster than state-of-the-art solvers, with the gap growing as problems get bigger - only about 20 prompts in. Big caveat: he hasn't released it yet, so nobody has independently verified those numbers. He says he's holding it for the big SAT competition in April. His takeaway: "This is how all future progress will be made: through human-AI symbiotic feedback cycles." If even part of this holds up, a lot of research timelines just got compressed.
18
THIS PLANE NEVER EXISTED. NEITHER DID THE BOAT. OR THE OCEAN SPRAY. Over 780,000 views on a clip of a massive passenger jet skimming just above the ocean, roaring over a tiny boat full of people filming on their phones - and according to the poster, it's AI-generated with Seedance 2.5. The breakdown they shared: - A small boat spots a huge airliner approaching low over the water - The jet passes directly overhead - Its engines blast an enormous cloud of white spray off the surface - Then it's gone, leaving a trail of mist behind The low camera angle and the tiny boat sell an insane sense of scale. A year or two ago, AI clips gave themselves away in seconds. Now people have to be told it's fake. That's the new reality: if it looks impossible, assume it might be generated - and check before you share. Would you have believed this was real?
9
"IT MARKS THOSE CHIPS UP 800 TO 900%." NYU finance professor Aswath Damodaran - the man Wall Street calls the Dean of Valuation - puts Nvidia's money machine under the microscope, and his take is going viral. His breakdown, in his own words: - Nvidia designs the chips, TSMC manufactures them - Nvidia owns none of the factories - "It marks those chips up 800 to 900%." - A typical chip company runs ~40% operating margins. Nvidia? ~70%. And then the warning: "The 70 cannot stay 70 because you're going to get tensions on both ends." His reasoning: the biggest buyers don't want to keep paying that premium. "Meta and Microsoft are working on their own chips because that way you don't have to pay the premium." The hottest chip company of the AI era, and one of the most respected valuation minds says its margins have a ceiling. Is the Nvidia premium about to crack - or does AI demand keep it alive?
21
THE ROBOTS AREN'T COMING FOR DESK JOBS FIRST. THEY'RE COMING FOR THE FLOOR. A clip racking up nearly 100K views shows what the poster describes as a robot laying laminate flooring - the kind of hands-on trade work people spend years getting good at. For years the comfortable assumption was simple: AI takes the office jobs, and skilled manual trades stay safe because the physical world is too messy for machines. Clips like this are why that assumption is getting shakier by the month. Think about the pitch any robot maker can make to a contractor: no back pain, no sick days, potentially round-the-clock operation. Once the quality matches and the cost drops, that's a hard argument to ignore. Is it ready to replace a crew today? Nobody can tell that from a 30-second clip. But the direction of travel is obvious. Flooring today. What's next?
1
12
AMAZON IS REPORTEDLY ABOUT TO DROP $7.5 BILLION ON ONE AI STARTUP. Per the Financial Times, Amazon is in advanced talks to buy Decart, a San Francisco AI startup founded in 2023 - and the deal could come together as soon as early next week. Why would Amazon pay that much for a company that's only a few years old? According to the reporting, Decart's tech helps AI models run more efficiently and squeeze more out of the chips they run on. The FT frames it as a move to help AWS compete with Nvidia in the race for computing power. That's the real battlefield right now: not just who builds the smartest model, but who can run AI the cheapest at massive scale. Nothing is signed yet - these are advanced talks, not a done deal. But $7.5 billion for a startup founded in 2023 says everything about how hot AI infrastructure is right now. The compute wars just got a lot more expensive. teapawt.org/zerohedge/status/21241…
*AMAZON CLOSES IN ON $7.5BN DEAL TO BUY AI START-UP DECART: FT *AMAZON DEAL COULD COME TOGETHER AS SOON AS EARLY NEXT WEEK: FT
15
AI ISN'T JUST A TECH STORY ANYMORE. IT'S HOLDING UP THE US ECONOMY. According to The Kobeissi Letter, AI-related investment added +1.27 percentage points to real US GDP growth from Q1 through Q3 2026. That's roughly 65% of the economy's entire ~2% cumulative growth over that stretch. Where the money went: - Software: +0.45 pts - Computers and peripherals: +0.37 pts - Communication and electrical equipment: +0.21 pts - R&D: +0.16 pts - Data center structures: +0.07 pts Read that again: nearly two-thirds of US growth traced back to the AI build-out. Chips, servers, software, data centers - this is the biggest capex wave of our lifetime, and it's showing up in the national numbers. The question nobody can answer yet: what happens to growth if that spending ever slows down? teapawt.org/KobeissiLetter/status/…
BREAKING: AI-related investment contributed +1.27 percentage points to real US GDP growth in Q1 through Q3 2026. This represents ~65% of the economy's cumulative growth of ~2% over this period. Software investment contributed the most, at +0.45 percentage points, followed by computers and peripheral equipment, at +0.37 percentage points. Communication and electrical equipment contributed +0.21 percentage points, R&D +0.16 percentage points, and data center structures +0.07 percentage points. AI investment has become the backbone of the US economy.
5
HOLLYWOOD SHOULD BE NERVOUS. A 2.5-minute AI-generated video is blowing up right now - and the creator says it was made with Seedance 2.5, ByteDance's video model. The creator's verdict: "This AI video is absolutely amazing" - and "Malfoy is crazy." Think about that for a second. Not long ago, AI video meant a few wobbly seconds of melting faces. Now people are dropping multi-minute clips built around famous movie characters, and tens of thousands of people are watching. Fan films. Alternate scenes. Entire "what if" storylines. All from a prompt and a laptop. The tools are here. The only question left is who gets creative with them first. Watch it and judge for yourself:
1
45
A $40 BOARD. TWO EVENINGS. A PIXEL WATCH-STYLE INTERFACE AT 60 FPS. Google reportedly paid $40 million for Fossil's smartwatch tech and $2.1 billion for Fitbit. Meanwhile one developer grabbed a cheap round-screen board and vibe coded a smartwatch UI that, by his own account, feels close to an Apple Watch. The setup, as described in the post: - Waveshare ESP32-S3 board, 466x466 round AMOLED, no GPU - Vendor libraries thrown out - no LVGL, no Arduino_GFX - Custom display driver, touch driver and drawing library written from scratch with Claude Code - Google's open Wear OS samples ported in: rolling digits, Pixel Watch card gallery, weather, a music player The claimed result: 60 frames per second, zero dropped frames, no tearing - and zero frames drawn when nothing moves. On the stock firmware, the icons reportedly tear apart mid-scroll. The best part? It's open source on GitHub and you can flash it from Chrome in a couple of minutes. Hardware makers, this is your wake-up call. Video: PopaSpinka on YouTube
1
73
A 200-POUND-CARRYING HUMANOID BUILT FOR THE JOBS HUMANS DREAD. Foundation Future Industries just pulled the cover off Phantom MKII - its second-generation humanoid aimed at construction, manufacturing, mining and even defense. The numbers being shared are wild: - Up to ~90 kg (200 lb) payload while walking - 29 degrees of freedom - IP67 weather protection - rain, snow, dust - 3 kWh battery, roughly 4 to 6 hours of runtime - Field-replaceable modular joints - 1.83 m tall And the brain? Cortex II - an AI system designed to learn physical tasks by watching humans demonstrate them. Foundation also says it cut parts cost roughly in half vs the first model, with production planned for early 2027. Optimus, Figure, Atlas... the heavy-duty humanoid race just got another serious contender. Would you trust this thing on a job site?
23
CLAUDE CODE HAS A SECRET SWITCH... AND IT LETS THE AI WRITE ITS OWN PROGRAMS TO DRIVE YOUR TOOLS. It's called Code Mode - an experimental feature from Anthropic, and the official announcement has racked up 2.4M+ views. Here's why people are hyped: Normally the agent calls one tool at a time - Bash, Read, Edit - waits for the result, then decides the next move. With Code Mode, Claude writes a small JavaScript script that chains your tools together, running multiple commands in parallel, passing arguments along and filtering outputs - so far less raw junk floods the context window. The killer bit: per Addy Osmani's breakdown, long-running jobs like test suites or big builds can run as background processes in auto or bypassPermissions mode - and Claude can keep answering you while they run. Want to see exactly what it generated? Ctrl+O expands the collapsed Code rows in the transcript. To try it: CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1 Agents are going from clicking buttons one by one... to writing their own programs.
Check out Code Mode, a new, experimental way for Claude Code to work with tools. Claude writes scripts that chain your tools together in sophisticated ways, and it can answer you while long commands are still running. Opt in with CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1 code.claude.com/docs/en/code…
1
25
AI CAN NOW FAKE YOUR CUSTOMERS... AND THE ANSWERS COME OUT SCARILY CLOSE TO THE REAL THING. A post about this research just blew past 2 MILLION views - and the method is free and open source. The paper: "LLMs Reproduce Human Purchase Intent via Semantic Similarity Elicitation of Likert Ratings" (PyMC Labs, using Colgate-Palmolive survey data). The trick is called Semantic Similarity Rating (SSR): - Don't ask the AI "rate this 1 to 5" (you get bland, middle-of-the-road answers) - Give it a customer persona, show it a product concept, and let it answer in its own words - Then map that text onto a 1-5 scale with embedding similarity The test: 57 real personal-care product surveys, 9,300 human respondents. The result: synthetic consumers hit roughly 90% of human test-retest reliability - and they explained WHY they rated things the way they did. Market research that takes weeks... simulated with LLMs. Paper: arXiv 2510.08338 Code (Apache-2.0): github.com/pymc-labs/semanti…
1
31
THE ROBOTAXI STOPPED. THE HUMAN DIDN'T. This clip out of downtown Austin, Texas is going viral - 158K+ views in a few hours. According to the person who filmed it, a Tesla Cybercab robotaxi came to a stop... while the human driver nearby kept on going. His take: "360 awareness is amazing." No steering wheel. No pedals. Just cameras and AI making the call at night in a busy downtown. Clips like this are exactly why the robotaxi debate is heating up: every near-miss caught on camera becomes a referendum on who you'd rather trust behind the wheel - a person, or the machine. Watch it and judge for yourself.
36
AI CHATBOTS MAY BE JUDGING MEN MORE HARSHLY THAN WOMEN. FOR THE EXACT SAME THING. A research paper called "Harsher on Male?" built GAMA-Bench: 1,298 gender-mirrored conflict scenarios, from intimate relationships to public social disputes. Same misconduct. Same context. The only thing flipped: whether the person was a man or a woman. Then they measured how the models handed out punishment, empathy, escalation, instruction and blame. The finding: when the actor was male, the LLMs got more punitive, escalated more and assigned blame more directly. When the actor was female, the framing turned more therapeutic and empathetic. And it didn't go away. Per the paper, the pattern held across model families, scenario types, model scale - and even with explicit thinking-style reasoning switched on. People ask AI for advice on relationships and conflicts all the time. If the scales are tilted, that matters. Paper: arXiv 2606.14068
14
UPLOAD A FILE. GET A SHELL. THAT'S HOW BAD THIS ONE WAS. Security firm pwn.ai just dropped "Upload2Shell" - a single Ghostscript vulnerability (CVE-2026-125396) that turned everyday file handling into remote code execution. We're talking file uploads, email attachments, even PRINT JOBS. Their write-up lists paths across Debian 13, Google, WordPress, Roundcube and more. The scary part: per their research, almost every app they tested with file uploads could be tricked into passing a file with a trusted extension to Ghostscript. Something as routine as cropping an image or generating a preview could expose it. The AI twist: they say GPT 5.6 Sol, working with researcher Paulos Yibelo, found the bug. The fix is out - Ghostscript 10.08.0 or your distro's backport. If your servers touch untrusted files, PATCH NOW. AI is now hunting zero-days. Defenders, keep up.
Good morning, we are releasing Upload2Shell - where you upload a file and get a shell. We wrote about a single vulnerability that turned file uploads, email attachments, & even print jobs into remote code execution paths across Debian 13, Google, WordPress, Roundcube, etc...
2
157
AI IS NOW MAKING PHONE CALLS FOR YOU... AND PEOPLE CAN'T STOP LISTENING. This creator says he had his Grok Bot call The Home Depot to ask one simple question: are fresh-cut Christmas trees in stock? It didn't stop there. Per the creator, the AI even asked about sizes and prices - the kind of follow-up questions you'd expect from a real person on the line. He posted the full call. 361K+ views and climbing. Think about what this means: no hold music for you, no phone menus, no awkward small talk. You just tell your AI what you need... and it dials. The era of AI agents doing your errands over the phone is already here. Would you let an AI make your calls?
16
YOUR AI PROMPTS MAY NOT BE AS PRIVATE AS YOU THINK. OpenRouter - the hub devs use to route requests to tons of AI models - just confirmed it found data-handling problems with some of the inference providers on its platform. Their own words: security researchers found the issues, and there was "no breach of OpenRouter itself." But the response says a lot: - 3 providers OFF-BOARDED - Many endpoints pulled from ZDR (zero data retention) eligibility - Affected customers notified - The researchers' testing being built into ongoing checks How it was reportedly caught (per a viral 328K-view breakdown): a researcher planted secret links in prompts sent to MiniMax models... and someone opened those links 40 to 130 hours later. That breakdown also claims AtlasCloud, Novita and StreamLake were forwarding requests to MiniMax. If you send sensitive data through ANY AI router, check who actually handles it.
26
YANN LECUN JUST TOOK A SLEDGEHAMMER TO THE LLM HYPE. One of the godfathers of modern AI is saying out loud what the industry does NOT want to hear. His point: when you train an LLM on human-produced knowledge, you are training it to accumulate what is ALREADY known. And in his words, that's not intelligence... that's memory. Rote learning. Not reasoning. Not invention. Not creativity. Not adaptivity. Not the kind of intelligence we see in animals. Billions are being poured into scaling these models - and one of the most respected names in the field is calling it memorization. Is he right, or is he about to be proven wrong? Watch the clip and decide.
17
OPEN SOURCE AI IS UNDER FIRE... and David Friedberg just drew a line in the sand. His warning is brutal: restricting open source AI is "a path to tyranny." His argument: any group that controls what people see, access and are allowed to do with AI - and sets the "weights" on history, science and "the truth" - ends up with tyrannical power, even if it never meant to. The fix he lays out: regulate, restrict, even criminally penalize what people DO with software... but don't limit the software itself. Because labeling and locking down open models, he says, creates a "regulated monopoly" for closed source products - and if those get aligned with government agencies, they get power over YOU and every business on Earth. "None of us should want that. I don't care what your political affiliation is." Open vs closed is THE fight of the AI era. Which side are you on?
18
THE TERMINATOR JUST WENT ON A LATE-NIGHT TALK SHOW... AND ROASTED ALL OF HUMANITY. This AI edit has rocketed past 1.9 MILLION views on X. The setup: the legendary "El Risitas" laughing interview meme, but the guest's head is swapped for a glowing red-eyed Terminator endoskull, the host is a robot too, and the studio audience? Rows of Terminators. The subtitles are pure gold: "lets accelerate AI, right? ... oh no! AI is dangerous! we need to slow down. I mean: which one is it?" "first they pump trillions into our development... and suddenly they wonder why they're facing mechanical gods" "we even warned them what would happen in movies" And the closer: "my IQ is over 40.000 and I'm immortal" It's a joke. But it lands because it's poking at a contradiction everyone in tech can recognize right now.
1
63