The Kalashnikov concern and Russian MOD have elected to test a new round. The 6.02 x 41 round. This round will improve upon existing 30 caliber and 21 caliber Russian rounds. They are introducing a new rifle for the military. Provisionally called the AK-22. This will be a continuation of the AK platform. The new round, 6.02 x 41 mm, surrounded by the 5.45 x 39 and 7.62 x 39. The provisional 6mm round is intended to defeat NATO body armor. Projectile mass 103 to 116 grains, 800 meters/sec (2,625 ft/sec). Three rounds... 6.7 x 45, 6.5 x 39 and 6.02 x 41 Note the absence of cannelures. Russian designers have rejected the use of cannelures to break, creating a pseudo-hollow point effect. They engineered more towards penetrating body armor. US 6mm SAW (squad automatic firearm), developed for ballistic equivalent to the M193 at farther ranges. This was developed in the 1960s and early 1970s. 105 gr, 2520 ft/sec, 1,488 joules at the muzzle. Range, out to 800 to 1000 meters. Peak pressure of 45,000 psi. Last photo - proposed 6mm SAW. An M16 variant was also proposed. A little trivia - the 6mm SAW was developed with computer models. The round's peak pressure is lower than the M193 and much lower than the 277 Fury.

Thursday, 4:40pm Slack: “Can you turn this by Monday?” Your SOW said two weeks. If you say yes and invoice the same flat fee, you just donated a weekend. New Premium lesson in Classroom → AI That Pays: how to put a rush multiplier in the proposal before anyone asks, the exact email/Slack reply when the late request hits, and why inventing the number after you already said yes feels like a bait-and-switch. Name the 1.5× / 2× line while they’re still deciding. Then the Thursday reply is just reading the menu.

There's a new course in the Premium classroom called Build AI Systems: From Model Design to Custom PC. It has six video lessons, each with a full written walkthrough underneath, and it follows the whole path from how a model gets designed to the machine it runs on. The first two lessons are the software side. You'll see how a model is put together for a real job (collecting examples, holding back a test set, and choosing between prompting, RAG and fine-tuning), then how to wire an automated workflow with a trigger, an LLM call, logging, retries and a spending cap. The other four are hardware. Lesson 3 explains why VRAM decides which models you can run, using the actual Ollama download sizes for Qwen 3.5, Gemma 4 and gpt-oss. Lesson 4 is two AI workstation parts lists priced for late September 2026. Prices are rough right now because memory makers are prioritizing AI data center orders (a 32 GB DDR5 kit is running about $400 to $560), so the lesson also shows where you can save money and what each swap gives up. Lesson 5 walks through building a regular office or gaming PC step by step, and lesson 6 covers first boot, stress testing, installing Ollama and measuring tokens per second on your own machine. It's aimed at members who use ChatGPT or Claude and want to know what's happening underneath, and at anyone who's pricing a PC for local AI or getting ready to build their first computer. The course is open to Premium members here: https://www.skool.com/ejm-future-tech-innovators-6708/classroom/42f82754 If you were putting together a machine for local AI this year, what would you mainly want it to run?

Canadian ingredient company Crush Dynamics just hooked an AI digital twin onto the fermentation tanks that turn leftover grape pulp into flavor powders and purées for food makers. Timothy Inklebarger at FoodNavigator covered it on Sep 22. Crush Dynamics, based in Summerland, British Columbia, already ferments winemaking leftovers into grape purée and powder that help brands cut sugar and sodium. The new partnership with AI developer Atomic47 Labs watches industrial sensors in real time, runs simulations, and tells the plant how to tweak the run instead of waiting on daily lab samples. Protein Industries Canada is putting $607,000 into the $1.4 million project; Crush Dynamics and Atomic47 cover the rest. Crush’s R&D director Michelle Cordeiro said the setup is expected to cut energy use about 30% once the models are dialed in. They’re also testing cherry, cranberry, apple, and avocado leftovers on the same platform. This is factory and lab tech, not a countertop gadget. You’re not buying it for your kitchen — food manufacturers are. If the energy and consistency numbers hold after scale-up, it’s one more path for waste streams to become ingredients without loading more sugar and salt into the formula. Anyone here already spotting fermented grape powders on ingredient lists?