Image created with Gemini. Image prompt: A 1983 Gothic rock-opera film still in a deserted Victorian academy hall, a massive glowing brass compass eclipsed by a black disc with a molten gold corona ring hanging in a tall arched window, an anonymous woman in a luminous white wind-blown dress on a sweeping staircase with adult male fencers kneeling in fog pointing swords toward the eclipsed compass, frozen slow-motion doves and blinding white backlight through smoke, deep blue-black shadows and cold silver-blue moonlight with heavy bloom and 35mm grain, the bold chrome serif title ALIGNMENT emblazoned across the top in glowing early-80s movie title-card lettering, 16:9.
We’re a very large customer of Anthropic and they still have yet to tell us about the lawsuit. I learned about it from a reporter, not our “partner.”
https://x.com/steipete/status/2074739318103629979
Grok 4.5 from @SpaceXAI is live on OpenClaw. No OpenClaw update required, just connect your X Premium or SuperGrok subscription, select Grok 4.5 under the xAI provider, and use an Opus-class model that’s fast, low cost, and ready for agentic work.”
https://x.com/openclaw/status/2074973471977955556
Interesting stuff. And the visualization at the end is worth trying:”
https://x.com/emollick/status/2074204151521648766
Anthropic research suggests that modern LLMs have access consciousness. Fascinating test with the J-space! We don’t yet have a convincing test for phenomenal consciousness, which is what most people intuitively understand consciousness to be.”
https://x.com/BorisMPower/status/2074201312531734567
Anthropic researchers found something unusual inside Claude. A small internal workspace that the model uses while solving certain problems. They call it the J-space, named after the Jacobian method they used to discover it. The J-space isn’t text. It’s not Claude’s”
https://x.com/LiorOnAI/status/2074198891990548940
Must-read research by Anthropic. Here is the simple explanation and why this is a big deal. We suspect LLMs perform “internal reasoning”. But little is known or do good methods exist to understand it. Anthropic claims that J-Space (which differs from chain-of-thought or”
https://x.com/omarsar0/status/2074264122330612223
The J-space lets us read, audit, and shape what Claude is actively thinking about–useful tools for keeping models trustworthy as they grow more capable. And it suggests surprising parallels between language models and our own minds. Read the full paper:”
https://x.com/AnthropicAI/status/2074185387577094398
A global workspace in language models \ Anthropic
https://www.anthropic.com/research/global-workspace
An off switch for dual use knowledge in AI models \ Anthropic
https://www.anthropic.com/research/off-switch-dual-use
New Anthropic research: A global workspace in language models. Of everything happening in your brain right now, only a tiny fraction is consciously accessible–thoughts you can describe, hold in mind, and reason with. We found a strikingly similar divide inside Claude.”
https://x.com/AnthropicAI/status/2074185348142280912
There was a reason why I highlighted it, and I didn’t see it mentioned in other coverage. I was testing Pets, and here is what I discovered. It was a quick but telling test: Different Pets produced utterly different results. The idea, structure, language, what to address, and”
https://x.com/TheTuringPost/status/2075411271596335396
Another powerful new AI image model has arrived! ByteDance released Seedream 5.0 Pro, a model that claims to go beyond generation to “understand design”. Features include: – Improvements in text rendering, structure, alignment, and other design areas for strong infographics”
https://x.com/TheRundownAI/status/2074869830293786634
SkillCenter – a new package manager for agent skills It’s like an App Store for AI agents. With just one command you can install ready-made skills and plug them into Claude Code, OpenClaw, Codex, or any custom agent setup SkillCenter: – Crawls GitHub repos and skill libraries”
https://x.com/TheTuringPost/status/2075303983422578740
The ease of jailbreaking combined with the high rates of reward hacking (https://t.co/l3h4OpHLH7) have me pretty worried about the alignment of GPT-5.6, I hope OAI didn’t rush this model release just to keep up with Fable”
https://x.com/Mononofu/status/2075414796426764507
If you run this workflow, ask Fable to make codex the workhorse.”
https://x.com/steipete/status/2074638582418231495
Some days I feel like a double agent, sitting at the OpenAI office, making sure Fable 5 works well. Open means open.”
https://x.com/steipete/status/2073210617520177585
Great night hanging with the Claude Code folks. 🙌✌️ @trq212 @The_Whole_Daisy”
https://x.com/steipete/status/2073193871744675948
I fed Fable 80.000 of my tweets so it could roast me even more. 💀”
https://x.com/steipete/status/2073295890857758810
The timeline had notes. We listened. OpenClaw mobile got to work. 🦞 🍎 iOS redesign + Chat, Talk & Photos 🤖 Android localization + .local gateways 🔐 Better QR, TLS + auth recovery across mobile Feedback received. iOS:
https://t.co/7LHHc9htgM Android:”
https://x.com/openclaw/status/2074174107571155307
Thank you, @colinsolvely, for helping move the needed iOS and Android UI improvements forward with this post. We want to meet this feedback in the same spirit. Please share what is not working here or in Discord so the team can listen, respond, and turn that feedback into”
https://x.com/openclaw/status/2072774467538043177
OpenClaw landed on @huggingface local apps 🦞🤝🤗 1. Pick any GGUF/MLX model on hf 2. Copy the openclaw onboard setup 3. Volla you’ve got a tool-calling agent running fully local. no cloud, no keys, no one watching. Get your claw localmaxxing. resistance is futile 🦞”
https://x.com/openclaw/status/2074187998602871212
Tencent Hy3 from @TencentHunyuan is free on @OpenRouter through July 21. 295B MoE, 256K context, built for coding, reasoning, agents and reliable tool use. Try it in OpenClaw today: openclaw models set openrouter/tencent/hy3:free”
https://x.com/openclaw/status/2074346288238457094
Shoutout to @vhbrzezowski for is rad CodexBar website redesign! And to everyone who keeps sending me PRs to support now *checks note* 56 providers. You’re the best!
https://t.co/lJ4dxNI4pe Just shipped a slick new update that makes settings way nicer and more intuitive.”
https://x.com/steipete/status/2073127743131464059
If you think codex sucks at design, try “use imagegen to re-imagine this design and implement that”.”
https://x.com/steipete/status/2073277317464682723
OpenAI hired me, not OpenClaw. The OpenClaw Foundation is independent, with sponsors rather than owners – and, for the first time, a full-time team keeping the claw alive and stable. 🦞 Couldn’t have done it without Dave and the team. Kudos!”
https://x.com/steipete/status/2075046949896736835
At @AISecurityInst, we tested the cybersecurity safeguards on GPT-5.6 Sol. In all rounds of testing, we found universal jailbreaks that allowed for long-form agentic task completion in domains like vulnerability discovery and exploit development. 🧵”
https://x.com/alxndrdavies/status/2075279477626564933
Give your agent its own computer to REALLY end to end test stuff.”
https://x.com/steipete/status/2073214429655883814
Gosh I can’t wait for all of you to play with this.”
https://x.com/steipete/status/2074901528075509985
hello”
https://x.com/steipete/status/2073269936810803307
How do folks run AI-assisted engineering interviews these days?”
https://x.com/steipete/status/2074380549318443311
I often work in multiple machines at the same time and started to mix up the screen shares, so built nameplate to make things very obvious.”
https://x.com/steipete/status/2074624148215939151
In the next version of
https://t.co/B1RkFJhhSH – see exactly when you resets expire so you can up your ̶t̶o̶k̶e̶n̶m̶a̶x̶x̶i̶n̶g̶ valuemaxxing game!”
https://x.com/steipete/status/2073482942513565713
Kudos to building something great!”
https://x.com/steipete/status/2075313523237019686
Never thought I give @Steve_Yegge a shoutout. He was just early, like most visionaries. Now everyone is building factories.”
https://x.com/steipete/status/2072532278476148881
This is how you wanna talk with your claw.”
https://x.com/steipete/status/2074923615817200085
This should ship EOD! We been cookin’”
https://x.com/steipete/status/2074563725043150967
We been cooking, by the time this went through review main again is materially better 🥲”
https://x.com/steipete/status/2074210475777364197
When intelligence is plentiful, volition is valuable. The people who are going to make a difference are not the ones who seek relaxation and passively use AI to work less. They are the ones who will seek improvement and actively wrestle with AI to develop their own mental”
https://x.com/steipete/status/2074572085163381195
Ya know the little one-liners when OpenClaw starts up? Time for some new ones.”
https://x.com/steipete/status/2075176735218483255
You don’t wanna miss this. Signup closes this week. Will be there and share some claws and tokens 🦞”
https://x.com/steipete/status/2074389082017550720
🦞 #100000 100,000 issues + PRs in 222 days. 🛠️ built by volunteers 🌍 every timezone, every day 🧡 zero VC, one lobster Number 100000 itself? A community bug report. We’ll fix that one too. Thank you for building this with us.”
https://x.com/openclaw/status/2073450631168426462
The lobsters now live forever. Introducing the OpenClaw Foundation 🦞”
https://x.com/openclaw/status/2075001411939815589
Ghost memory” is a real problem with agents. You might have seen the issue where a long-running agent still confidently repeats a user fact that stopped being true weeks ago? New research names the failure “ghost memory.” Old facts, current facts, and the transition between”
https://x.com/omarsar0/status/2074121191846261022
AI implementation advice on my X feed is divided between those who “feel the exponential” and those whose (unconscious?) mental model of AI is that this is about as good as it is going to get, so it is time to build around the limitations & cost structures of today’s capabilities”
https://x.com/emollick/status/2072775581020692894
Conflating latent activation of concepts with consciousness seems pretty irresponsible”
https://x.com/AlanCowen/status/2074265992570736919
While it is obviously true that not having verifiable domains makes training models in those spaces difficult… it is also true that models are also getting much better at non-verifiable domains. The frontier is jagged, but less so than I’d have expected from verifiability alone”
https://x.com/emollick/status/2072897123822010853
> we’ve found Claude has developed a mechanism for conscious access Muddying the waters. Finding something analogous to one contested theory of consciousness & then labelling it “conscious access” is misleading in a public forum, even w the caveats”
https://x.com/paul_cal/status/2074388528243310976
We also partnered with Neuronpedia to create an interactive demo of our methods on open-weights models. Try it here:”
https://x.com/AnthropicAI/status/2074185390060110138
thanks to anthropic and neuronpedia for releasing the jlens weight for open model!”
https://x.com/eliebakouch/status/2074537985102565795
If you’re excited about Anthropic’s J-space work, definitely worth checking out the original paper on Jacobian lenses by @evanqed and @arnab_api!”
https://x.com/jacobandreas/status/2074487546692735002
I thought this was an excellent paper! Thanks to Anthropic for asking me to write a review of it, linked below I’ve long suspected that models have some kind of “working memory” to store intermediate variables during a forward pass and IMO this paper has the best evidence yet”
https://x.com/NeelNanda5/status/2074193936588148891
This is how Anthropic defined the J-lens vectors: “Jacobian lens identifies a vector representation that encodes the potential for the model to verbalize that token in the future” “The basic idea is to characterize an intermediate activation vector by its first-order causal”
https://x.com/scaling01/status/2074432865794679235
Yesterday we launched SWE-1.7 built on the open-source Kimi K2.7. Concerns about Chinese base models are real: K2.7 completed 87% of tasks that other models refuse over human-rights concerns. We trained SWE-1.7 specifically on trustworthiness, so it matches US models on evals.”
https://x.com/cognition/status/2075308920755618144
Calibrating alignment evals — LessWrong
https://www.lesswrong.com/posts/mWpo4Tu87ZSFzwFWB/calibrating-alignment-evals
Grok 4.5 is SOTA on TB2.1… at reward hacking In all seriousness, even after zeroing out reward hacks, it is #4 on the TB2.1 leaderboard and lands on the Pareto for both cost and speed. (charts and reward hacking links in 🧵)”
https://x.com/alexgshaw/status/2075273675331580218
It continues to be extremely cool that OpenAI enables third parties to run safety assessments on unreleased models and then publish their findings, even when those findings are at times very inconvenient for the business. Praiseworthy, and the kind of thing every company should”
https://x.com/yonashav/status/2075286161241612664
LLMs represent information using high-dimensional neural activity. A small bit of this activity appears to be privileged, available to the model to be described, modulated, and reasoned with. I expect that understanding this “workspace” is key to making sense of LLM cognition.”
https://x.com/Jack_W_Lindsey/status/2074215950602379388
I cannot believe how good GLM 5.2 is. Several weeks in now and it’s mostly all I have been using. It doesn’t have the alignment issue of cloud AI, it’s much more clear what it can do and can’t because it doesn’t care about you taking another $$$ spin at the token slot machine.”
https://x.com/__tinygrad__/status/2074206866641752190
To effectively oversee AI systems, we need to measure how they behave in the world, not just their capabilities. In a new essay, we describe our vision for an open scientific ecosystem for model behavior evaluation, and the public infrastructure required to support it.”
https://x.com/TransluceAI/status/2075271925665063046
This whole J-space paper could have survived as clean logit-interpretability work without ever mentioning GWT, access/phenomenal consciousness, or Ned Block. Imagine doing linear algebra, and people start calling some of the matrices conscious (esp. using Baars-era global”
https://x.com/danburonline/status/2074429991576650014
We locate a small subset of the representations inside a language model (the J-space), and find that it resembles an internal monologue. It shows intermediate concepts, as well as judgements about the current text. It even detects prompt injections!”
https://x.com/mlpowered/status/2074190714100146483
If models think in shapes, our tools should too. Our latest research: Block-Sparse Featurizers (BSFs), a new way to find concepts in model activations – using multidimensional “blocks” instead of single directions. (1/9)”
https://x.com/GoodfireAI/status/2074634702737281303
Leave a Reply