Day 4 highlights: Glen Pass (the steepest climb) and subsequent descent. Felt like a different planet. https://t.co/4XCWUfhASS

Day 5 highlights: all of the switchbacks 🙃. We missed a bear cub by maybe a minute. https://t.co/AaP0J5GPMe

Moral of the story: it was gorgeous. Highly recommend surrounding yourself with friends who push you, for fun. https://t.co/MVRwBfyLZV

+2 more

P.S. first stop off trail: Wine https://t.co/jOJcpPERYJ

Hot chili off the stove, cornbread from the oven, and kick-off in 30. It is well with my soul. https://t.co/172GziMpwX

Can you guess the most predictive day of the year for $AAPL stock well into the next year? It’s not Christmas. & it’s not Black Friday. It’s Apple’s annual keynote event…a bona fide LikeFolio holiday. SPOILER ALERT: 2022 is the Year of the Apple Watch https://t.co/eHYzvY1F5E https://t.co/jYFLcmEhRV

Just received my 4th aggressive text message promoting a conservative candidate in less than 24 hours. Which one of you signed me up for this?! https://t.co/OowePzvt1n

Sharing some positive airline feedback! Mia turned 11 today and got the biggest surprise thanks to a @SouthwestAir flight attendant. She’s obsessed with their snack mix and when my aunt asked for an extra bag to take home for her… they over delivered. Birthday made 🙌🏻 https://t.co/2O7GAF6rLo

Presenting your Titans of Tub and hulking, hefty heroes! These bears will soon duke it out to see which tanking tummy reigns supreme. Check out their bios and pics at https://t.co/V2bF4olcpu. Tell us who you think should win (extra points for being punny)! #FatBearWeek https://t.co/ptkXQcLyMW

Deep in enemy territory. Y’all pray for us. https://t.co/yQA39tF7gy

@J_B_Horne We just finished this in my book club and honestly it was my favorite read of the year. If you like non-fiction and magazine-style writing, it's a hit. Bonus: learned a lot about wine -- AND there's a trading story in there... https://t.co/Fak9aMwQ3T

Ultimately, we believed subscribers were going to be key for this event...and luckily this is where our data shined the most. Netflix subscribers were outperforming in a big way. https://t.co/EACfeAyGtG

#AI threatens entry-level jobs in Australia as university grads struggle to get hired #RiseoftheRobots https://t.co/ruYzG94agj

2. Traditional metrics of engineering hours/days don’t work for AI. Two metrics I use to evaluate the complexity of a project: - how many instructions I need to give AI until it completes a project - how often I have to interrupt it because it goes into the wrong direction Across my projects, the interruption rate is about 1 in 4 instructions. This means I still need to actively monitor the agent, rather than letting it operate autonomously.

2. Tool use Given how easy it is to add new tools (e.g., with MCP servers), the number of tools a user adds to an agent can explode. Too many tools make it easier for the agent to choose the wrong action, making them dumber. They caution against removing tools mid-iteration. Instead, you can force an agent to choose certain tools with response prefilling. Ex: starting your response with <|im_start|>assistant<tool_call>{"name": “browser_ forces the agent to choose a browser. Name your tools so that related tools have the same prefix. Eg: browser tools should start with `browser_`, and command line tools should start with `shell_`

Our findings: 1) For short seqs, increasing density or size provides gains. For long seqs, high sparsity performs best. 2) Higher sparsity is possible for decoding and larger models. However, most configs deteriorate performance significantly for at least one task. 3) There is no single sparse attention method that excels across all tasks. 4) We establish scaling laws based on model size, sequence length, compression ratio, and task id that reliably predict performance.

Trump Secures $1.35 TRILLION Trade Victory Over EU 🇺🇸 While the media obsesses over distractions, President Trump just redefined global trade. ✅ $750B in U.S. energy exports ✅ $600B in new EU investment ✅ 15% tariff on EU imports — ZERO on U.S. exports ✅ NATO freeloading ends ✅ EU dependence on Russian gas? Crushed. This is economic warfare won through negotiation. Europe blinked. America wins. Trump didn’t just make a deal. He rewrote the rules of sovereignty, trade, and power. The result? •Energy dominance •Massive manufacturing boom •Strategic deterrence without a shot fired EU leaders stood frozen as Trump laid out the terms—with zero wiggle room. Ursula von der Leyen? Speechless. Tariffs work. Leverage matters. The deal-maker of all time just reminded the world why America leads. This is what real leadership looks like. America First. Globalist era over. 🇺🇸

Drawing Day - Get your tickets online at https://t.co/Fff8vgqRdJ . You could win #HALF of a projected $50,000. Drawing tonight at 8:30! Lic#Org2527 We will be at South End BBQ Watch LIVE on our Queen of Hearts LIVE FB page https://t.co/0j74COI5I6

Drawing day - Get your tickets! Go to https://t.co/BbJjbIA21S for a chance at #HALF of a projected $14,500. Drawing tonight at 6. #ShirleysWay Lic#Org2527 Watch LIVE on our Joker's Wild LIVE FB page https://t.co/s4NBhvVla4

@GregLangdon You might want to check out the @jonerp podcast on practical AI, I think you'd enjoy our discussion on the subject https://t.co/qApC5QaHYu

@Stummer__Schrei Lol I tried to remove it but I couldn't manage to https://t.co/1L1P9d0AH1

We want to explain why we don't provide regular updates. Our development team is working tirelessly on the game, and as it's a singleplayer adventure, we're cautious about not spoiling the experience prematurely. Furthermore, with the high level of attention our project is receiving, any new feature we develop is at risk of being replicated. It's flattering, in a way, to see elements of our vision being echoed in the industry, even if only in a superficial form. However we're introducing a lot more of new innovative elements, which is why we've chosen to keep these developments under wraps for now. What we have publicly shown of the game is just a tiny part of what we are doing. Rest assured, the development is progressing smoothly, we're just opting to work more discreetly and focus on the quality of the game. But expect some big reveals this year. We're aware of other games that may try to be similar to us, but we're focused on our unique vision and not in a race with others. Our silence doesn't mean inaction, rather, we're diligently working to bring something truly special to you. We can't wait to share it with you when it's ready. Unrecord has not been released yet and is still under active development. There's no beta or early access available at this stage. Please be aware, any claims to the contrary are likely scams, viruses, or simply ripoffs. Stay vigilant and safe. Remember, our goal isn't to rush the game for quick profit. Thank you for your patience. Also we've climbed to #12 on Steam's most wishlisted games list with nearly 1,200,000 wishlists! Your enthusiasm is overwhelming. Thank you! We're excited about the future and believe you will be too.

+1 more

🛠️ @isaac_flath's productivity system EXPOSED 😲 ⚡ UV scripts + Just files + AI agents = instant content generation 🤔 Simple beats complex. Personal beats universal. Speed beats perfection. 👇 https://t.co/sbZQI4JGd3 https://t.co/pJbIkiIhEr

I really like Isaac's workflow, he uses a Just file (it's like Make but better) to expose a bunch of tools to his coding agent Seems way cleaner and more maintainable than MCPs for personal productivity (less indirection) https://t.co/o2lpwrKY0R

2. Inference Run inference on pretrained models by the DEIM authors. https://t.co/KWC70PqWiG

Run batch inference https://t.co/JVmfpBV3ug

3. Training Instead of multiple config files, you can train on a custom dataset with just the following Python codes https://t.co/85AKWFXedk

Here are some sample results I got by training on customs datasets on @roboflow Universe https://t.co/C6QTaOMlXt

+1 more

Monitor your training with Tensorboard. Compared to the original DEIM repo, I've added more metrics into the logs such as - mAP values at different iou - F1 score - Recall scores - Precision scores - PR curve plots These are incredibly helpful to debug your model and dataset https://t.co/XmjYMdC87J

4. Export Export the trained model https://t.co/1DKKcOxQzS

The original DEIM repo supports ONNX export but it only exports the model. DEIMKit takes it one step further by exporting the model and the preprocessing into the ONNX graph. Full credit to @PINTO03091 for the implementation! https://t.co/0oPCgNNuJ2

This means, at inference time you can pass in the image at any resolution, and the ONNX model will resize it automatically into the size that the model expects. This has 2 benefits. - Less custom code on for ONNX inference. - Inference time speedup https://t.co/pnz1wCbKmC