nycdatasci

Dupe: https://news.ycombinator.com/item?id=47247645

The idea of stateful models/interactions in an enterprise is extremely powerful. Is anyone aware of open source projects that have a similar goal? I'm looking for stateful conversations, with collaborative agent/skill refinement.

To head off the semantics debate: I don't mean a model rewriting its own source code. I'm asking about 'process recursion'—systems that analyze completed work to autonomously generate new agents or heuristics for future tasks.

"We demonstrate that our IsoDDE more than doubles the accuracy of AlphaFold 3 on a challenging protein-ligand structure prediction generalisation benchmark, predicts small molecule binding-affinities with accuracies that exceed gold-standard physics-based methods at a fraction of the time and cost, and is able to accurately identify novel binding pockets on target proteins using only the amino acid sequence as input."

It seems like a key challenge here is not just creating a protein that will bind to a specific site, but also ensuring that off-target binding won't happen. Is this feasible? I'm not familiar with this space, but RefSeq [1] shows 442M proteins and the human protein atlas seems to only cover 17.4k [2]. Do we have comprehensive knowledge of human proteins that would allow us to identify off-site affinities?

[1] https://www.ncbi.nlm.nih.gov/refseq/

[2] https://www.proteinatlas.org/

Since many posts mention lack of substance, providing a link to the All-In Podcast from last week in which they discuss Clawdbot (prior to re-brand). https://www.youtube.com/watch?v=gXY1kx7zlkk&t=2754s

For the impatient, here's a transcript summary (from Gemini):

  The speaker describes creating a "virtual employee" (dubbed a "replicant") running on a local server with unrestricted, authenticated access to a real productivity stack—including Gmail, Notion, Slack, and WhatsApp. Tasked with podcast production, the agent autonomously researched guests, "vibe coded" its own custom CRM to manage data, sent email invitations, and maintained a work log on a shared calendar. The experiment highlights the agent's ability to build its own internal tools to solve problems and interact with humans via email and LinkedIn without being detected as AI.

He ultimately concludes that for some roles, OpenClaw can do 90%+ of the work autonomously. Jason controversially mentions buying Macs to run Kimi 2.5 locally so they can save on costs. Others argue that hosting an open model on inference optimized hardware in the cloud is a better option, but doing so requires sharing potentially sensitive data.

The landing page for the demo game "Voxel Velocity" mentions "<Enter> start" at the bottom, but <Enter> actually changes selection. One would think that after 7mm tokens and use of a QA agent, they would catch something like this.

I don't think this is a relevant comparison. Snap guns break pins, which isn't the case with this robot.

Is this from 2024? It mentions "With global data center demand at 60 GW in 2024"

Also, there is no mention of the latest-gen NVDA chips: 5 RNGD servers generate tokens at 3.5x the rate of a single H100 SXM at 15 kW. This is reduced to 1.5x if you instead use 3 H100 PCIe servers as the benchmark.

Work around from comments:

  rm -rf ~/.claude/cache
  mkdir -p ~/.claude/cache
  echo "# Changelog" > ~/.claude/cache/changelog.md
  chmod 444 ~/.claude/cache/changelog.md

Have you experimented with all of these things on the latest models (e.g. Opus 4.5) since Nov 2025? They are significantly better at coding than earlier models.

Text-to-Image Models Leave Identifiable Signatures: Implications for Leaderboard Security

https://arxiv.org/pdf/2510.06525

There are no visible watermarks, but model makers can use steganographic codes to identify outputs from their own models.

The arena concept doesn’t work for image models due to watermarks.

Not rock, but silver-plated and EM-shielding: https://vollebak.com/products/shielding-pants

From Semrush’s website: “ Why You Need Keywords You need keywords because they are the exact words and phrases people type into Google. When you know the keywords people use to search for things related to your website or business, you can create pages and content that answer real searches and attract interested visitors.”

$1.9B isn’t a huge sum but a business focused on keywords for SEO? Isn’t this like buying a pet dinosaur right before the K-Pg extinction event?

You seem to be suggesting that current frontier models are only trained on text and not "sensor data". Multi-modal models are trained on the entire internet + vast amounts of synthetic data. Images and videos are key inputs. Camera sensors are capable of capturing much more "sensor data" than the human eye. Neural networks are the worst way to model intelligence, except all other models.

You may find this talk enlightening: https://simons.berkeley.edu/talks/ilya-sutskever-openai-2023...

I ordered one that arrived last week. It seems like a great idea with horrible execution. The UI shows strange glitchy/artifacts occasionally as if there's a hardware failure.

To get a sense for use cases, see the playbooks on this website: https://build.nvidia.com/spark.

Regarding limited memory bandwidth: my impression is that this is part of the onramp for the DGX Cloud. Heavy lifting/production workloads will still need to be run in the cloud.

"When it comes to software engineering at scale, nothing beats Google"

I agree with many of your statements, but this one simply isn't true. They've really struggled to integrate AI into products in a useful way. Do you remember glue on pizza? Founding Fathers reimagined for DEI? The Brain/DeepMind merger was largely an acknowledgement of their many misses from a product perspective.

What makes TikTok fun is seeing actual people do crazy stuff. Sora 2 could synthesize someone hitting five full-court shots in a row, but it wouldn’t be inspiring or engaging. How will this be different than music-generating AI like Suno, which doesn't have widespread adoption despite incredible capabilities?

I assume many here are familiar with the album and the story. For the uninitiated, this is "the best-selling piano recording and the best-selling solo album in jazz history".

The story behind the album is worth reading about: 11:30 PM start time, broken piano, tired pianist who almost bailed, sold-out show, etc. The entire piece is improvised and Jarrett starts by playing the same melody that the opera house played to let people know to take their seats.

You can listen here: https://www.youtube.com/watch?v=Pd_Kti6jvy8&list=OLAK5uy_lap...

“ This is the first step toward a more useful ChatGPT that proactively brings you…”

Ads.

How are you handling the auth token/captcha on the website? I'm not a legal expert, but my impression is that automated solving of captchas during a scrape carries higher risk. That said, amazing project!

The attack surface for agents with MCP access grows exponentially with the number of tools. On the scale of thousands of tools, I think it's nearly impossible to understand potential interactions and risk. Do you have any innovative controls in place that would help a CISO get comfortable with a product like this in an enterprise context?

Maybe someone here can archive? I did a quick search for it, and landed at: https://australia.hackclub.com/slack/

…Which renders upside down. Maybe an Australia joke? The primary server appears to be at slack.hackclub.com

You can find a graph showing the relationships between investors and entities here: https://staging--atlantic-council-spyware.netlify.app/

The headline can't be taken at face value. "Largest" is based on the number of investing entities (including individuals), not something more objective like dollars invested. Also, the US is not making these decisions as the headline implies.

The ruling: https://storage.courtlistener.com/recap/gov.uscourts.dcd.223...

I've used products from topazlabs.com for the same problem and have generally been happy with them.

If you want to see many more than 50 words and also have an appreciation for 3D data visualization check out embedding projector (no affiliation): https://projector.tensorflow.org/

If the goal is to get drunk, there are lots of cheap ways to do it and it's fine to buy high-fructose-corn-syrup-based "mixers". If you're looking for great cocktails, consider the PDT (Please Don't Tell) book and/or app by Jim Meehan. When his recipes call for lemon juice, you'll be buying and squeezing (do not buy "fresh squeezed lemon juice"). If you want more of a sensory experience, take recipes from the Aviary in Chicago (they also sell a book). Prepare to spend several days preparing each drink.

Just take the win. It seems like such a bizarre thing to nitpick like this with a prior employee that has voluntarily opened up to you and agreed with many of your points.