echo.hive

My longest running consulting client(2+ years) wanted to build an AI chatbot for his restaurant Today he's the Lead of GenAI R&D at Amsterdam University of Applied Sciences. Because @WonderWhyAI is relentlessly driven and curious, our monthly 1-on-1 sessions went deep: - Full-stack AI application development - Agentic systems and retrieval architectures - Real-time voice AI - The bigger questions: what's worth building, where AI is heading, and how to position yourself ahead of it Working with Fausto has been one of the highlights of my consulting work. If you're building with AI, I'd love to do the same for you. Right now I'm offering 30% off for a limited time (Link in comment) 1-on-1 Consulting (1 or 3 hours/month) A dedicated thinking partner for whatever you're building with AI. Architecture decisions, technical deep-dives and strategy. we work on your problems.

At 750 TPS, this would have taken 24 minutes

You know, there is non-zero chance that this doesn’t stop

i saw a few people on this app saying "you have to review the ai-written code" i havent seen them in a while. presumably they have been stuck reviewing a single day of codex work for the past few months

There seems to be a real risk of instruction pollution in GPT-5.6 Sol from leftover instructions and skills. In fact, I noticed Sol is eager to use all the skills available when it gets the chance. In some cases, this is a plus, and in others, a distraction. That's why I cleaned up my agents.md and removed all third-party Codex skills and plugins. I'll add back only those my projects actually need. The release of 5.6 is a good time to do a bit of housekeeping.

@daniel_mac8 But what if there is a bubble and ASI is cyclical What then? I ask you 😝

Sir Demis Hassabis has a brand new article on the "Dawning of a New Age" and Frontier AI. Demis has no doubt that major existential changes are coming to humanity. He proposes a regulatory body similar to FINRA, with formalized standards to regulate frontier labs and frontier model development. From the article: "We will realise we were standing in the foothills of the singularity - nothing less than the dawning of a new age for humanity." Well worth a read.

Seriously tho: Years ago, CNNs naturally evolved edge detectors and layers of representations like the human visual cortex We found Jennifer Aniston neurons in early GPTs like it happens in the human brain Then LLMs started speaking just like humans like humans do Then we found emotional representations in the model’s latents like “despair” which can lean the model towards cheating as could happen in humans Then we found J-space, the “global workspace” or “access consciousness” analog like humans have How many more things like this will we need to find…

@usr_bin_roygbiv I told it: complex is fine where necessary but complicated is not Let’s see how that goes

sol is a genuine 10x from 5.5 xhigh it's creating subagents for entire projects to orchestrate teams of agents to create a state machine for the script it's going to write for checking an email

@nicdunz @flowersslop On a more positive note, I would like to congratulate you on your good prompt today I feel the same sometimes, especially lately when I realize 5.6 ultra was creating an entire civilization from first principles when I just asked for a tribal hut village

@flowersslop anyways

its so satisfying when you enter flow state and write a god tier prompt to codex

I never see anyone post about Claude design anymore. what’s interesting to me is that I find that environment incredibly powerful for conceptualizing more than actually building. Just chat with Opus within Claude Design. ask them to help you pull references and build a mental model of whatever you’re working on. it’s a complete game changer. I promise. I’ve never witnessed a better source of *deep* cut references in my life. it’s crazy, actually. haha. opus will pull like 10 *still frames* from obscure movies and commercials to articulate what they are going for visually. and they’re usually spot-freaking-on. blows my mind. anyway. treat Claude design like a chat sometime. that’s the moral of the story. you’ll have fun. <3

@NickADobos this is soo true!

One big problem with codex usage limits is I now run both a bunch of personal life chores and tasks with codex. So if I vibecode too close to the sun I brick my day planner & workout log for a few days. So I end up being too conservative to keep my small personal infra alive and don’t run ambitious loops to limit max and test codex

@usr_bin_roygbiv like next month probably

@hive_echo wdym later?

what if it is later realized that 5.6 way of overdoing style coding was actually the "real programming" and what human programmers manually did until now was actually "vibe coding"

After Codex temporarily removed its 𝟱-𝗵𝗼𝘂𝗿 𝗹𝗶𝗺𝗶𝘁, I kept more sessions and subagents running at once. My Mac bogged down. My Mac is an 𝗠𝟭 𝗠𝗮𝘅 𝘄𝗶𝘁𝗵 𝟲𝟰𝗚𝗕 𝗼𝗳 𝘂𝗻𝗶𝗳𝗶𝗲𝗱 𝗺𝗲𝗺𝗼𝗿𝘆. Lower-memory machines will hit memory pressure and swap sooner under the same load. I asked Sol to map the full process tree. It checked for stale dev services, duplicate MCP servers, headless browsers, login items, VPN and network extensions, browser helpers, and updaters. I cleaned up the redundant items in batches, rebooted, used the Mac normally for a day, then checked again. The same multi-agent workload now 𝗿𝗮𝗿𝗲𝗹𝘆 𝘀𝘁𝘂𝘁𝘁𝗲𝗿𝘀, and I can keep more sessions open. OpenAI can improve inference and quota accounting. I still had to clean up stale processes on my machine. I turned the audit into a cross-platform prompt for macOS and Windows. The first pass is 𝗱𝗿𝘆-𝗿𝘂𝗻 𝗼𝗻𝗹𝘆. Full prompt below. 👇

Robot Lorentz look a bit startled

Country of geniuses in a Solvay

@swyx @threepointone @sama Ok but, why not DONTs: Durable Object, Nonlocal and Trans-temporal

@threepointone @sama crazy that DOs exist in closed and open source and is still not everywhere as the default.

hmu @sama

@amuldotexe @BenjDicken @dejavucoder @joy014 The distribution of performances of 5.6 family itself is a reduced 2d manifold representation of a high dimensional complex dynamical system that is the shadows on the Plato’s cave 😝

@hive_echo @BenjDicken @dejavucoder @joy014 this is helpful -- it shows its not linear

Sol is getting all the attention but Terra is actually insane value for the $.