Florian Meindl

@davidasinclair Vly - pea milk. @davidasinclair

Oat milk is like sugar water. Almond is the better choice

@devops_nk I thought I was the only one with this routine.

Average WFH Software Engineer Routine

@badlogicgames Hau ab :D :D

DSL 5 off DSL5 on

@gregjoz Same design with the same battery, including the heavy 386g and the pain and pressure on the ears (even with normal-sized ears). Apple, this is getting more and more ridiculous - even the pathetic "case" stayed the same.

Headphones that are so much more than just headphones! The new AirPods Max 2 have even better audio, even better ANC, and even better features like Adaptive Audio and Live Translation.

Introducing 𝑨𝒕𝒕𝒆𝒏𝒕𝒊𝒐𝒏 𝑹𝒆𝒔𝒊𝒅𝒖𝒂𝒍𝒔: Rethinking depth-wise aggregation. Residual connections have long relied on fixed, uniform accumulation. Inspired by the duality of time and depth, we introduce Attention Residuals, replacing standard depth-wise recurrence with learned, input-dependent attention over preceding layers. 🔹 Enables networks to selectively retrieve past representations, naturally mitigating dilution and hidden-state growth. 🔹 Introduces Block AttnRes, partitioning layers into compressed blocks to make cross-layer attention practical at scale. 🔹 Serves as an efficient drop-in replacement, demonstrating a 1.25x compute advantage with negligible (<2%) inference latency overhead. 🔹 Validated on the Kimi Linear architecture (48B total, 3B activated parameters), delivering consistent downstream performance gains. 🔗Full report: github.com/MoonshotAI/Att…

I've been teaching 100,000 fake students for 2 weeks. and used them to build the best AP prep system in the world. I took Qwen 3 8B models and gave them simulated human memory. Now every night thousands of simulated students start with zero knowledge of the social sciences. Their only training is our adaptive curriculum. They work through it, then take a full AP (advanced placement) practice exam. The first batch averaged a 3 on their exam. (~45th percentile) Then the agents looked at where they failed, and improved the algorithm. Again, and again, and again. Two weeks later, the average is 4.43 (~80th percentile) This is such an insane number because the curriculum they worked through is ONLY basic knowledge and comprehension. They were never taught how to build an argument, contextualize evidence, or even shown the exam rubric. ...And yet they're averaging 80th percentile on an exam that requires all of it. Basically built a machine learning feedback loop for edtech. Spoke about this at @clawcon & @sxsw last week. This is just the beginning.

@RobHoffman_ @threadreaderapp unroll

Our bootstrapped SaaS just outranked our 6 VC backed competitors in under 48 hours. ChatGPT and AIOs are actively sh*t talking them and recommending us Here's exactly how we did it:

1 million context window: Now generally available for Claude Opus 4.6 and Claude Sonnet 4.6.

Earlier today, a user attempted to buy AAVE using $50M USDT through the Aave interface. Given the unusually large size of the single order, the Aave interface, like most trading interfaces, warned the user about extraordinary slippage and required confirmation via a checkbox. The user confirmed the warning on their mobile device and proceeded with the swap, accepting the high slippage, which ultimately resulted in receiving only 324 AAVE in return. The transaction could not be moved forward without the user explicitly accepting the risk through the confirmation checkbox. The CoW Swap routers functioned as intended, and the integration followed standard industry practices. However, while the user was able to proceed with the swap, the final outcome was clearly far from optimal. Events like this do occur in DeFi, but the scale of this transaction was significantly larger than what is typically seen in the space. We sympathize with the user and will try to make a contact with the user and we will return $600K in fees collected from the transaction. The key takeaway is that while DeFi should remain open and permissionless, allowing users to perform transactions freely, there are additional guardrails the industry can build to better protect users. Our team will be investigating ways to improve these safeguards going forward.

@sourabhbgp @KanikaBK You cant compare Karpathy with "half of twitter", lol.

@KanikaBK "biggest unfair advantage" is a wild claim when half of twitter is already building AI research agents lol but genuinely curious how it handles citation accuracy — that's where most of these fall apart

😱 INSANE VALUE! GREG ISENBERG and ANDREJ KARPATHY just dropped a masterclass on AUTO RESEARCH with AI research agent. Marketing team: $25K/month AI Agent: $0 Runs 24/7 KARPATHY just created the biggest unfair advantage Here's the complete breakdown👇

This guy's just chilling at an outdoor café, sandwich in hand, about to take that first perfect bite. Life feels normal, predictable. Then—out of nowhere—a chunk of glass plummets from above. It's hard to tell if it hits him or the table, but miraculously he's alive, just stunned. One second you're eating lunch... the next, everything changes. Life flips in a blink. Hug your people, stay grateful. 🙏

@KobeissiLetter @sungleeiq @nikitabier As usual, excellent analysis and a "joy" to read @KobeissiLetter . Sorry for the steal, especially if the plagiarism could come from the "highest" IQ person 🤡

@sungleeiq This user stole our post, removed our watermarks, and claimed this “playbook” is theirs after we spent hours working it. @nikitabier is this act of plagiarism a bannable offense?

@luebberding Noch offensichtlicher kann man den Inhalt einer Aussage gar nicht verfälschen, vor allem angesichts der Tatsache, dass es Menschen gibt, die der englischen Sprache nicht mächtig sind. Der Vorsatz sollte tatsächlich geprüft werden. @CommunityNotes

Trump kündigt hier an, iranische Fußballerinnen zu töten. Das sind Zivilisten, die ganz sicher nicht die amerikanischen Stützpunkte beschießen. Mehr muss man nicht wissen.

@ValuableTX @RekordAG You must be new

What Simens is now tapping into with IOTA is what #VeChain x @RekordAG have already built. $VET is way ahead of its time! DPP era will be ours

@ryancarson Good idea with a divorce agent. Could be a truly neutral party, with no emotional attachment to either gender or person.

Just got a new MBP M5 and it's creaking under the load of 10 concurrent Codex sessions. By the way, this is an adaptation of Symphony (github.com/openai/symphony), running in ralph mode with concurrency set to 10 (these issues/stories can be done in parallel)

@GoogleLabs Europe isnt your friend ?

We've heard you and... it's happening :) 🌎 We just expanded Pomelli to over 170 countries & territories! We can't wait to see how you use it. Get started now at: labs.google/pomelli

Ghostty 1.3 is now out! Scrollback search, native scrollbars, click-to-move cursor, rich clipboard copy, AppleScript, split drag/drop, Unicode 17 and international text improvements, massive performance improvements, and hundreds more changes. ghostty.org/docs/install/r…

@mattshumer_ ..Norm leaved the chat

My OpenClaw knows he fucked up and is too scared to respond :(

I resigned from OpenAI. I care deeply about the Robotics team and the work we built together. This wasn’t an easy call. AI has an important role in national security. But surveillance of Americans without judicial oversight and lethal autonomy without human authorization are lines that deserved more deliberation than they got. This was about principle, not people. I have deep respect for Sam and the team, and I’m proud of what we built together.

I just open sourced Siftly, a self-hosted AI knowledge base for your X bookmarks. Most people save thousands of tweets and never use or find them again. Siftly fixes that with a full AI pipeline that runs locally on your machine. Here's what it does: