Lotto

@HououinTyouma @shinboson Hermes agent + qwen 27b could do it

@shinboson uploaded whole genome of me and my ex girlfriend and asked ChatGPT to grow our child's brain in a vat and connect it to play animal crossing, pokemon, and doom

this is a dual-use psychotechnology and it must be regulated now

@jhmonteiro Woah that’s an esoteric stack

@LottoLabs I'm a simp with 4070 mobile 8Gb. OminiCoder-9b 4q_k_m ggfu full VRAM 192K ctx FTW! 35t/s. Rock solid.

Qwen 27b remains my favorite still, gonna do a write up on all the models tested so far

@sergeykarayev Meh get me a phat card and a couple instances of qwen27b and docker then tell me that’s not the future

Running agents locally is a dead end. The future of software development is hundreds of agents running at all times of the day — in response to bug alerts, emails, Slack messages, meetings, and because they were launched by other agents. The only sane way to support this is with cloud containers. Local agents hit a wall quickly: • No scale. You can only run as many agents (and copies of your app) as your hardware allows. • No isolation. Local agents share your filesystem, network, and credentials. One rogue agent can affect everything else. • No team visibility. Teammates can't see what your agents are doing, review their work, or interact with them. • No always-on capability. Agents can't respond to signals (alerts, messages, other agents) when your machine is off or asleep. Cloud agents solve all of these problems. Each agent runs in its own isolated container with its own environment, and they can run 24/7 without depending on any single machine. This year, every software company will have to make the transition from work happening on developer's local machines from 9am-6pm to work happening in the cloud 24/7 -- or get left behind by companies who do.

@Forgework_ Aren’t you a business already 😂

@LottoLabs Been working on this as well! Excited to see how it goes!

I’m gonna start a business w/ Hermes agent + 27b and see how far we can get, I bet we can have an autonomous business built in a day or two

Hermes agent + qwen 3.5 27b on a 3090 All local, your data stays with you, runs 24/7, create tools and reporting to run autonomously while you sleep, access through telegram, WhatsApp, discord etc., simple set up, avid group of technical people developing it, mlops focused but expandable to any needs, unless you’re needing sota coding this stack covers you for everything and more.

Tell me why I need Claude or a clawd bot? I don’t think I need more than grok currently which is a search engine on steroids. Convince me otherwise and I’ll send you $500 in bitcoin.

@lyc_aon 🥹✌️

@LottoLabs I can't abide

Burner computer w/ hermes, tailscale, host llm locally (27b or api) All the perks, not forced into platforms or payments, physically airgapped You need less not more

@iamgoncaloalves Llama.cpp server/lmstudio server

@LottoLabs Are you running on llama.cpp, ollama or other?

@lyc_aon Trust me brother

@LottoLabs burner computer? fuck that we do it live

@yacineMTB We need the great agent browser condom more than ever

"we are going to stop bots on x" Refresh the timeline 90% LLM generated anxiety bait from posters outside of the western world Sweet

@duganist I use my local desktop to serve the 27b and have hermes on a separate dev computer then ssh into it w/ tailscale or use telegram

@LottoLabs I'd like to host 27B on a remote node and have Hermes tag into it instead of it needing to be local or an approved API model. Or SSH in a terminal and run Hermes in that terminal, I suppose is fine.

@danveloper Join the club

I actually hate MoE's now. Not just because they're difficult to hardwaremaxx, but it's actually a really dumb architecture (no offense to anyone). They naively approximate a graph without any of the benefit of graph traversal. We're sending a blind person down a path and we've trained something to nudge them onto a different path to get to the end, but it doesn't know the next part of the map until the person has walked down the street. I hate this.

@jdmsec Depends what your doing, it’s not going to replace opus but if you have a defined scope and prompt it can work well

@LottoLabs hows its orchestration? opus 4.6 is expensive w Hermes, but afraid Im going to lose massively on orchestration/reasoning if I move to a local model

99% of people would be fine w/ qwen 27b and Hermes agent The 1% can offload work to sota

@francophilll Yes

@LottoLabs hermes over openclaw?

@user351513 @agentsdotmd That’s why it reminds me of sonnet 3.5

@agentsdotmd @LottoLabs You can't just give it a general mission and 500 files and have it figure everything out for you. But my theory is that with the right tooling and prompt curation you can get close enough for most work.

@ChemPhysMajor @johnhanacek @sudoingX Nice I’ll check out the new ones

@LottoLabs @johnhanacek @sudoingX I had issues with the original Qwopus at 27B. But Jackrong just released a v2 at 4B and 9B with way more Claude training data and actual benchmarks. The 9B v2 sounds like it's pretty much just as accurate but the thinking is effective at dramatically shortening the thinking.

Gonna try GLM 4.7 flash w/ Hermes + 3090 Probably similar to the other 30b a3b models but it is what it is Running out of models to fit on the 3090 w/ tool calls

@DoDataThings You using it in an agent harness or just normal chat inference?

@LottoLabs The Qwen team did really well on the 3.5 series. Enjoying it so far

@mweinbach Or small models that think less

@agentsdotmd *definitely not opus level knowledge

No definitely not but like for 99% of general population it’s enough, for devs it’s usable, it kinda reminds me of sonnet 3.5 where it wasn’t insanely smart but it was really sticky to the prompt and steered easily. I think it’s probably smarter than 3.5 but that’s the vibes I get and 3.5 was famous for a reason