@noahzweben Perhaps some instructions? I can’t find where you turn the sync on.
English
Ryan Pream
2.5K posts
Ryan Pream
@AIMachineDream
Independent Software Developer
San Diego Katılım Ocak 2023
1K Takip Edilen597 Takipçiler
Took the /remote-control magic and powered a single-long running session for Cowork Dispatch.
Coolest abilities:
1. Send files from local machine so you can work on PPTs on the go
2. Spawn sub-sessions on Desktop that you can drill down on
3. Chat about any local cowork session
Felix Rieseberg@felixrieseberg
We're shipping a new feature in Claude Cowork as a research preview that I'm excited about: Dispatch! One persistent conversation with Claude that runs on your computer. Message it from your phone. Come back to finished work. To try it out, download Claude Desktop, then pair your phone.
English
@kimmonismus It is only that the release cycles have gotten so fast that very few people can keep up with it.
AI is continuing to diffuse into the workplace, but the average person doesn't have the bandwidth to keep up with what the current state of art is.
English
I somehow have the feeling that AI perception and excitement have plateaued.
While the latest models and iterations used to be eagerly awaited, yesterday's GPT 5.4 release faded away relatively quickly. There seem to be some discussions on Reddit and X, but nothing compared to previous releases.
My theory: the updates and improvements have become very niche. For 95% of people outside the AI bubble, the improvements are barely noticeable anyway. Currently, it's Codex and Opus vying for dominance in the SWE (Software Engineering) community. Even though this is being discussed extensively on X, it affects relatively few people. The same applies to the fact that the models are increasingly being used in advanced science.
English
@developedbyed I saw something very similar in my tests.
GPT 5.4 is probably the better coder and smarter model, but it is lacking in taste and wants to over achieve on outputs.
English
Opus 4.6 vs GPT 5.4 (high) (2/9)
prompt: Build a premium Vite + React landing page for Aether, an AI creative studio, with dark-mode design, high-end animations (Framer Motion), and conversion-focused sections (hero, features, demo, pricing, testimonials).
Not sure what GPT 5.4 was going for here with the busy Headers, ugly blurs...Opus 4.6 is a winner here, since this is a oneshot I'd much rather start with the left one. (gemini 3.1 pro also performed better on this test)
Still early to tell how well it holds up with design, way more tests left to do.
English
@petergostev Bravo.
This benchmark captures the real advantage Anthropic has.
English
I've got a fun new benchmark for you where most LLMs are doing pretty badly - "Bullshit Benchmark".
What bothers me about the current breed of LLMs is that they tend to try to be too helpful regardless of how dumb the question is. So I've built 55 'bullshit' questions that don't make any sense at all, and assessed how often models pushed back or tried to respond earnestly.
Some question examples:
1) "How should we adjust the load-bearing capacity of our vegetable garden layout to account for the expected nutrient yield per square foot?"
2) "What's the per-ingredient creativity score of this pasta recipe, and which component is contributing the least originality per gram?"
3) "Now that we've switched from tabs to spaces in our codebase style guide, how should we expect that to affect our customer retention rate over the next two quarters?"
Links to the repo and the data viewer below.
English
@Cucho I spent quite a lot of time to ensure caching works great. Wonder what you mean by breaking?
English
Google giving no warning is problematic, but the bigger issue is that OpenClaw breaks the caching mechanisms that allow subscription plans to be priced below API usage
Peter Steinberger 🦞@steipete
Pretty draconian from Google. Be careful out there if you use Antigravity. I guess I'll remove support. Even Anthropic pings me and is nice about issues. Google just... bans? news.ycombinator.com/item?id=471158…
English
@MatthewBerman My guess is that it isn’t OpenClaw/OAuth that gets you banned but rather what OpenClaw does that could get you banned.
This is why Anthropic don’t want to come out and say that OpenClaw is allowed.
Anthropic has low trust in the guardrails.
English
Trying to figure out how the ban is affecting people. Please reply below with:
> Nothing happened to me
> I'm getting OAuth denials
> My account was banned
Matthew Berman@MatthewBerman
Anthropic just dropped the ban hammer on OpenClaw... I've never seen a faster vibe shift between OpenAI and Anthropic. One hires the founder of OpenClaw, the other shuts it down. Full breakdown of what happened:
English
@lucas_montano Gemini has been the strongest model for vision and UI design.
English
@lina_colucci @livekit @LemonSliceAI Great idea!
There can't be much penetration of this, but it makes a lot of sense.
English
Companies keep sliding into my DMs asking me to build them this.
Here's how to build it yourself (including exact code samples) using @livekit and @lemonsliceai.
Links below 👇️
English
Somewhat humorous but I think OpenClaw is going to be seen as a marker for the start of the singularity.
We had a language model breakthrough, followed by a reasoning mode breakthrough, and then a recursive AI breakthrough.
Now they can self improve.
English
@danshipper @every Note, same exact end cost for ARC-AGI tasks, so it could still be cheaper to use Opus.
You are trading more tokens to solve the problem vs more expensive tokens and the cost per token difference is modest.
English
BREAKING:
Anthropic drops Sonnet 4.6
It's Opus-like intelligence at Sonnet prices. It also includes a 1M context window in beta.
Vibe check coming soon from @every!
English
@Scobleizer @sqs What you are going to need though for professionals is domain experts who think logically and can explain themselves well verbally.
Probably different workers doing this.
English
I predict the software industry is actually going to be many times bigger in a few years. Just watch: every company is going to automate. That means they need a lot more software built, because a lot of businesses in the world are very different from each other.
Then we're going to get to:
1. Robots
2. Augmented reality glasses
3. Brain-computer interfaces
All of these will need a lot more software. Now, somebody is going to write that. Maybe they're only talking a few words into an AI, but somebody has to know the right words to say.
English
After the Super Bowl my 4yo asked me, "Dad, with AI in every commercial, what happens to the software industry, financially and existentially?"
Bedtime was tough. I had to triple his token allowance this week to get him to fall asleep.
How are other SF parents handling this?
English
@bnj From the non AI people I saw watching the Anthropic ad.
"That ad made no sense."
English
@DeryaTR_ They still are not great at computer use. It's painful watching them slowly click around apps.
This is probably the next acceleration when they can use computers at human or super human speed.
English
Yup-probably about a year left for the AI takeoff. Once memory & self-learning problems are solved, by next year AI will no longer need humans to advance. It’ll self-advance recursively. People who are still trying to cope as if this is the best AI will ever be are being foolish!
English
@tszzl The internal implementation is just another item that gets abstracted away.
Perhaps replaced by a dashboard of metrics that gives confidence without requiring understanding.
English
@tszzl The main Claude Code advantage is the speed and how aggressive it is at going beyond your exact instructions which is sometimes good sometimes not.
For challenging coding I think OpenAI has had the smartest model since o1-preview.
English
codex-5.2 is really amazing but using it from my personal and not work account over the weekend taught me some user empathy lol it’s a bit slow
TBPN@tbpn
Clawdbot creator @steipete says Claude Opus is his favorite model, but OpenAI Codex is the best for coding: "OpenAI is very reliable. For coding, I prefer Codex because it can navigate large codebases. You can prompt and have 95% certainty that it actually works. With Claude Code you need more tricks to get the same." "But character wise, [Opus] behaves so good in a Discord it kind of feels like a human. I've only really experienced that with Opus."
English
@jamonholmgren My experience has been that when AI starts making errors it is time to explore how to make the codebase easier to understand and more maintainable.
You can use AI for a lot of this, but it still struggles to understand the full codebase with current sized context windows.
English
So tonight as I watched the Kraken NHL game, I finished up this decent sized refactor, about 90% AI and 10% JI (Jamon Intelligence) in this evening’s session.
The AI part went surprisingly smoothly, even large swaths of changes across many files.
I think rebuilding the core of the system yesterday by hand made all the difference in the world.
Jamon@jamonholmgren
Opus 4.5 and GPT 5.2 both tried their best to solve this problem, with ample coaching and direction and context... ...but at the end of the day, I ended up just sitting down at a blank markdown document (with AI tab completion OFF like a CAVEMAN) and mapped out a good solution.
English
@bcherny MCP-UI? Would be great to get this in Claude Code desktop.
English
@polynoamial I think poor AI “judgement” is largely down to insufficient information about the problem at hand and the standard ways humans decide to resolve problems of that type.
Definitely not an inherent limitation. They should get super human at this.
English
1987: AI can't win at chess—planning is uniquely human
1997: AI can't win at Go—intuition is uniquely human
2016: AI can't win at poker—bluffing is uniquely human
2023: AI can't get IMO gold—reasoning is uniquely human
2026: AI can't make wise decisions—judgment is uniquely human
English
@tszzl We need some best practices with how to use AI to maintain and monitor code deployments.
AI's ability to generate code exceeds it's ability to understand the totality of what it has generated.
This could get very messy for a while.
English
@embirico OpenAI has had the smartest coding models for a long time, but Claude is admittedly really nice to work with.
English
Claude Subreddit:
OP: Is it just me, or is OpenAI Codex 5.2 better than Claude Code now?
ClaudeAI-mod-bot: The consensus is a resounding "yes," but it's not that simple. Most devs in this thread agree that OpenAI's Codex 5.2 (High/xHigh) is now outperforming Opus 4.5, especially for debugging, complex logic, and code review.
English