We trained an LLM trained on an LLM trained on a…🌀🌀🌀 If the original model is sycophantic or just 'weird', will those traits begin to amplify? Yes! But amplification is rare and typically comes at the cost of coherence—except in the case of DPO where things get dicey 🧵 https://t.co/Ax7yjt0LaX

Organize garments in the closet, including dresses that have been thoroughly wrung out. https://t.co/nLYNk8Z7K9

The @contra team published compelling research that disproves all of the AI doommaxxing being sold to creatives. Artists are making more art, earning more money, and staying independent. I’d argue the Creator Economy is stronger now than it was 5 years ago when the whole economy was dependent on Patreon.

Amazon suggesting a shirt that actually slaps. https://t.co/pl8qMemAjY

Even if you didn't get tickets to the sold out Bay Area premiere of I Love Boosters at Grand Lake Theater this Tuesday, April 28th- You can still come parry with us afterward at Crybaby!!! This is a real party. Not just networking. DJ Daghe is doing the music, so u know wsup. https://t.co/0zUzQlBSic

@vivianisvulgar Time for a reboot https://t.co/q2MVqPyJgt

@luci_holland One of my favorite history books written in the last 5 years It argues textiles are a foundational technology that shaped human history, driving innovation in science, economics, and culture https://t.co/hSLTEiuJv7

A study in April of this year found the jobs most at risk of being replaced by AI are bill collectors & telemarketers. https://t.co/oRaKQ5LTNS

Goldman Sachs has released a shocking report summarizing the job types that AI will take over. It states that AI has the potential to automate the work equivalent to 300 million people worldwide. Those particularly affected are knowledge workers, with ・Administrative and support: 46% ・Legal: 44% ・Architecture and engineering: 37% On the other hand, the impact on physical labor such as construction and maintenance is less than 6%. The result is that AI will first replace knowledge workers.

And that's how it all begins. https://t.co/GnYUZELuNa

And that's how it all begins. https://t.co/GnYUZELuNa

These are the moments that brought us to where we are. Watch new episodes of THE TESTAMENTS Wednesdays and stream THE HANDMAID'S TALE on Hulu. https://t.co/yRKOfbtPe2

The new DeepSeek-V4, like DeepSeek-V3, uses concepts from our 2024 paper on Self-Rewarding LMs -- see screenshots of their tech reports! (Congrats!) Classical :) Self-Rewarding LMs from Jan 2024: https://t.co/YZRuFphh0H https://t.co/KFfSNgbOLk

Fun fact - if you have a recent commit that mentions OpenClaw in a json blob, Claude Code will either refuse your request or bill you extra money. This is an empty repo, I'm just calling Claude Code directly. Insanity. https://t.co/JKthsztu1X

I am obsessed with this Japanese man using AI video to put himself into movies (he's on IG at @ai_am_furufuru) https://t.co/ePeVPkpEG4

Excited to introduce Gemma 4 Multi-Token Prediction Drafters⚡️Accelerated inference right in your pockets - Up to a 3x speedup - Same quality guarantees - Available in your favorite open-source tools https://t.co/x6WRgO4n9G

TIL that TeX versioning is based on successive approximation of pi, to showcase it approaching perfection... Currently it's on version: 3.141592653 Donald Knuth, the inventor of TeX, has stated that the "absolutely final change (to be made after my death)" will be to change the version number to π, at which point all remaining "bugs" will become "features" Pretty creative! (h/t @davidad)

AMD launches MI350P, its first PCIe "Instinct" in four years – packs CDNA 4 GPU with 4.6 PFLOPs AI compute, 144 GB HBM3E at 600W. https://t.co/uLAh7eokph

'I Am Your Father,' Reveals Trump To Horrified Mark Hamill https://t.co/jPM26ZxAYd https://t.co/UlvMcyvlSU

“Sparser, Faster, Lighter Transformer Language Models” LLMs are naturally sparse in their feedforward layers, but unstructured sparsity usually doesn’t get you real speed on GPUs, because the hardware stack is built for dense compute. The key idea of the paper is to redesign the sparse format and kernels around GPU execution, so FFN sparsity becomes practical instead of theoretical. With mild L1 regularization, this paper got >99% sparsity with little quality loss, and with gains in speed, memory, and energy.

Yeah okay, Lego bros, brodettes and brotheys are cooked with this one. GPT 2 Image can create full Lego sets! With actual Bricklink IDs so you can order the parts and build it. Whole new business opportunity here for the taking. https://t.co/4d4aJeEGTU

+1 more

Side quest: trying to write custom firmware to a DVD drive and get low-level hardware control. This has been a long-term AI bench of mine. Not quite there yet, but 5.5 made a ton of progress - it feels like we're quite close to the fun + scary world where any sufficiently complex device can probably be reverse-engineered and hacked with a bit of effort! Rough write-up of the current state: https://t.co/yIbRLMDstA

暇すぎてDVDのアレを再現するマシン作ってたら連休終わった.. 角に当たると爆裂レインボーに光る https://t.co/QlTobdhEFj

暇すぎてDVDのアレを再現するマシン作ってたら連休終わった.. 角に当たると爆裂レインボーに光る https://t.co/QlTobdhEFj

Microbial traffic jam. (Ciliates in biofilm, seawater sample from the Oregon coast, polarized light microscopy) https://t.co/ylDP9RyTeF

More tests with duckweed transformation and regeneration https://t.co/8gkCsuQozp

+1 more

Mildly obsessed with this bacteria growing on an old plate (cc @ContamClub ) https://t.co/lW2XDVSHJ6

+2 more

New EQ-Bench results! Opus 4.7: clean sweep, still the king. Deepseek 4: very strong, near frontier on EQ-Bench & longform writing. Kimi k2.6: Strong in shortform but seems to suffer degradation in longform writing. GPT-5.5: performs ~identically to GPT-5.4. https://t.co/slwFfDoBVj

+1 more

Newsletter update of last 2 weeks of AI/game/web/narrative links of interest... more splats and 360ies, web games, narrative and creativity, vis of Aaron Reed's Subcutanean procgen novel, and more! 1/2 https://t.co/vaHQM9gO48

@AhmedShahnab Some country files are being screwed by overseas territories https://t.co/8RRmMpROkO

I’m looking for the first full-time developer for https://t.co/nprGVKUWBn! You’d work with me on making new web projects and games for the site. Preferably in nyc If anyone is interested or has leads dm me or email hi@neal.fun! https://t.co/0bOvzn7UVc

🚨New preprint! We find evidence of LLMs enabling people to file lawsuits without lawyers (filing "pro se") at historically unprecedented rates in federal courts.👇 1/n https://t.co/JCj8oq5Jym