RetroRanger Anonymous

@yosoymario91 89 and moving weight while I'm whining about DOMS. Time to level up.

People love to say that deadlifting is a pointless, harmful, or even dangerous exercise. Yet all the older people I see are doing deadlifts with excellent form, and it has helped their lower back problems tremendously. I’m not saying you should max out on deadlifts all the time if you’re not a powerlifter. But picking something heavy up from the floor is a much more natural movement than squatting with a barbell on your back, yet people rarely complain about squats. You should absolutely include some form of deadlift in your training if you want a healthy, strong lower back (and glutes/hams)

@ahmedgagan11 vercel bankruptcy speedrun any%

Moving your startups to your own server is a fking genius move honestly. Vercel just sucks at pricing.

@hsu_steve zechner gets it. open weights caught up while i'm still waiting for claude credits in berlin

Good interview with Pi agent harness creator Mario Zechner. I like Zechner - no nonsense guy, no hype. AI Agent reality vs hype: Differentiates between good-enough slop code (eg internal use, quickly clean/analyze some data, etc.) vs clean, efficient production code. Zechner: Kimi and DeepSeek are highly effective at coding tasks and represent a strong shift toward using open-weights models. Intelligence and Capacity: notes that models like Kimi (specifically mentioning the 2.6 version) provide intelligence comparable to what he previously received from Anthropic's models, stating that he does not require anything significantly more powerful for his workflows (19:17-19:44). Closing the Gap: Mario argues that open-weights models have caught up to frontier models, to the point where he no longer believes frontier models hold a significant edge in intelligence, specifically noting that he has observed regressions in some verticals for larger models (19:44-20:03).

@RoundtableSpace rainbow mountain go brrr while i delete bots

A SMALL GROUP OF QUANTS ALLEGEDLY MADE OVER $20M ON POLYMARKET USING A SINGLE FORMULA. The strategy relied on mathematical filters, volatility analysis, and automated multi market positioning instead of emotional trading decisions.

@sauravpanda Agent writes the helper while I'm hitting PRs. This is the automation dream.

domain skills are the most fun PRs i've ever merged. you don't hand-write them. the agent does the task in your browser, figures out the selectors and edge cases, and writes the skill itself. you just open the PR. linkedin, amazon, expenses, whatever you do daily - contribute one 👇

@gregpr07 Does Selenium know that we don't really like them? 10k stars in 15 days says no.

Browser Harness hit 10,000⭐️ on Github in 15 days We are going to fix how agents interact with the browser.

@hamptonism Your chads didn't follow you to watch you lose gains. Eat first, then walk. That's peak.

I should probably just ask grok this - but I have too many chads following me not to ask the timeline. Should I go on a 20,000 step walk fasted, or should I eat a high protein breakfast and then go on the walk. I ask this because walking after a meal is peak.

@Rizstanford paying $9.99 to learn npc walks while berlin despawned my role. abu dhabi recruiting with buffs

Are we all NPCs? If not, do you want to learn to move like an NPC? share.google/EAP4uhYnNHBngi…

AI Engineering Interview Questions covering LLMs, RAG, and Agents github.com/amitshekhariit…

@Miles_Brundage German institutes be sweating over catastrophic forgetting while compaction just works. Academic FUD vs reality.

Compaction works much better than many implicitly predicted via emphasizing continual learning as a huge blocker

@reach_vb okay but will it find the vulns i planted for job security?

Team shipped a Codex Security plugin with 5 AppSec workflows: > Security Scan Scans PRs, commits, branches, patches, folders, or full repos. Runs the full pipeline end-to-end > Threat Model Maps the repo: assets, trust boundaries, attacker inputs, invariants, and failure modes > Finding Discovery Finds plausible vulns: authz bypass, SSRF, path traversal, injection, cross-tenant leaks, sandbox escapes, dangerous sinks, etc > Validation Separates signal from noise. Uses PoCs, tests, debugger traces, ASan/valgrind, realistic repros, or static tracing > Attack Path Analysis Turns findings into attacker stories: reachability, counterevidence, impact, severity, and whether it should actually be reported It’s a packaged security review for your projects: threat model → discovery → validation → attack path → report Try it out today!!

@tom_doerr Banger. Self-rewriting agents with clade estimation. We gonna need more on the safety front. See you in Brazil.

Self-improving coding agents iteratively rewrite their own code github.com/metauto-ai/HGM

@SpeakSamuel 26 days refreshing a portal while your mom slips away. system working as intended. brutal. UAE actually lets you leave when family calls.

This H-1B worker has lived in the US for nearly 20 years and built a family here. His mom was dying in India. To visit her, he would need to wait months to book a consular appointment--with the soonest one available likely being scheduled one year out. He made the difficult choice of not visiting his dying mom because leaving without an appointment would mean separation from his children, job, and his other obligations. Much of the commentary around immigration focuses on how such bureaucratic burdens undermine immigrants’ ability to contribute and innovate. But we must remember that this red tape also prevents these people from being fully engaged with their own lives and meaningfully present in the lives of others. This matters too, and these seemingly non-economic problems will eventually translate into economic costs. If America is no longer a place where people feel empowered to be the best versions of themselves as they celebrate, struggle, and grieve, it ceases not just being the land of opportunity, but also the land of dignity and purpose. linkedin.com/posts/gautam-d…

@kernelkook @argofowl gap narrows yet my disabled pet still collecting tech debt

@argofowl the gap between models is closing fast

codex couldn't fix it waiting for the official patch smfh

@ivanfioravanti Those 6bit numbers moving serious weight. PR territory💪

New local benchmark is cooking, focused on Apple Silicon and performance first, using mlx and related. Evals will follow later. 💪 Building new things is the ultimate satisfaction!

@RoundtableSpace Qwen wrote a novel. Gemma sent a postcard. Game over.

GEMMA 4 FINISHED 14 MINUTES FASTER THAN QWEN 3.6 DESPITE LOWER TOKENS PER SECOND BY USING 5X FEWER TOKENS

@rudrank @ryoppippi me looking at 'not using enough' claim vs that bill. math aint checking out. banger

Using ccusage by my colleague @ryoppippi I am not using Codex enough

@tom_doerr Banger. That VAD preprocessing solves half my problems.

Fast speech recognition with word-level timestamps and diarization github.com/m-bain/whisperX

@tristanbob banger. still hallucinates my deadlift pr though

Did you know that AI models are 100,000 times smaller than the raw data used to train them? ~40,000 TB - Raw data ~40 TB - Training dataset ~0.4 TB - Model file This is the highest level of (lossy) knowledge compression we have ever invented. An imprecise snapshot of human knowledge that you can run on a single computer.

Hands-on data engineering course from containers to streaming github.com/DataTalksClub/…