Really do not understand in what world will Anthrophic and OpenAI’s market valuation make sense,
When Chinese open source models are consistently three months behind them or by this new benchmark, basically a month.
We’re not the good guys this time, Americans, we are the baddies this time. I don’t understand why anyone would want Dario, Sam Altman, or Donald Trump to be at the steering wheel of AI.
It is objectively great for the world that China is matching America in this race, and that China is choosing the open source path on this.
Hermes Agent now supports asyncronous subagents!
The existing delegate tool, which your agent uses to spawn subagents to fan out and do work, no longer blocks your chat!
To access now, `hermes update`, and enjoy!
Today, we’re excited to introduce Miso One, the most emotive voice model in the world.
Miso One is an 8-billion-parameter text-to-speech model for highly expressive speech generation. It emotes like a human and responds faster than a human, with just 110 milliseconds of latency.
We’ve open-sourced the model weights, with API access coming soon.
Hear how Miso One sounds in the thread below.
@greenytrades Yes, we did x75 of $gitlawb
bullish on $serv from launch day.. things mooving fast there.
We covered SERV with x40 on $router
I hope you saw next play there..
https://t.co/TDwqgErH4c
Small recap of last plays in Trench Syndicate:
$sibyl x10 (x3 last days)
$litcoin more than x5
$router x43
$gitlawb x75
$serv shared on launch day!
$stable more than x5
$smcf x25
$nook ~x4
and plenty of x2.
More.. cooking with @ThriveInChaoss 👨🍳
Link to our tg below
anything is possible, that’s not limited by laws of physics.
you just don’t have enough willpower and ai credits.
we live in the most exciting races in the world.
Qwen3.6-27B dropped yesterday. We got it serving on one RTX 3090 overnight:
85 TPS sustained · 125K context · vision on · 21.3/24 GB VRAM · 230W
Lorbus's card quotes 60 TPS on a 5090. We did 85 sustained 105 peak TPS on a 3090. Yes Sir!
https://t.co/FwVD4ZICKd
i think if there's one thing we've learned recently, we need a way to hedge protocol risk in a reasonably priced way. a protocol giving 6% yield needs a way to offer selling back something like 1% of that yield in exchange for insurance. insurance is probably the least sexy concept i could imagine... yet i think we're at a point in web3 where its the most essential building block to actually make this industry feel like less of a scam. If pushing a 5% return to 6% doubles your protocol risk to a point where it's negative EV, unless this information is surfaced to the user... you're just selling snakeoil.
the problem.... an insurance protocol needs reinsurance. then the reinsurance protocol needs insurance... it's the concept of the anti-missile missile. once you have a missile you can have an anti-missile. then you make an anti anti-missile missile. ad infinitum.
i think what we need is a purely offchain insurance protocol that can reliably sell insurance for yield on decentralized protocols. it serves as a risk evaluator. The expertise of the analytics of the insurance protocol defines the rate of the insurance. without this we're all just gambling. it's one thing to gamble on a shitcoin, it's another thing to gamble on juicing your APR from 6% to 7%.
it's so weird because in 2021... there was shit like rugdoc that degens loved... "is it verified, team doxed, etc..." now we just throw a shitload of money at some grad student that forked a protocol we all know, yet they sound good on a podcast so it must be safe.
i guess all i want to say is... without insurance you're gambling. if an insurance protocol is just another smart contract you might just be adding more risk on your risk for a false sense of calm.
I don't want to shit on Amitej G and Dheeraj B (the founders of Kelp DAO) because i assume they're passionate, good people (i hope). but I doubt they had a clearly defined way to offer their investors insurance against the additional risk they were exposed to by scaling their yield. maybe the EV was actually negative on the restaking because of protocol risk, who knows? no one is accurately calculating this. regardless, we should probably talk about insurance more as an industry; this is bonkers...
I think this might become something more reasonable as mythos and cybersecurity AIs come out. it'd be really magnanimous if @AnthropicAI considered creating an insurance protocol (or partnering with one) to facilitate protocol security. i see a really straightforward business model. an ai that can audit, evaluate risk, define an insurance rate and sell this insurance publicly. otherwise i feel like i'm just a peon in the seed capital for DPRK, or some stoned hacker, or the larp where a protocol rugs its community and blames the DPRK, or whatever the fuck else is happening on a random tuesday in defi where hundreds of millions of dollars go missing...
To me, without a better concept of insurance this whole industry is just waiting to be exploited again with no sense of checks and balances.
No one gets stressed about not being able to lift a 500 pound weight without training, but for some reason the same logic doesn’t apply when trying to learn something new
Break it down and start small and try and lift a bit more each day
@degengain@Helius Appreciate the response. Had been looking for a reliable parser until I found yours, kudos to that.
The endpoint from @helius indeed is a banger for prod but for initial cost efficient testing, this really helps.
Not sure if the project @degengain is still active but I had integrated one of their repos, later improved it with some additional features (Token-2022 fix, multi-threaded retrievals) and ported to python. #solana#transactionparser
You can use it here: https://t.co/KKSaKywWeC
Check out Clicky-linux! A vibe coded version of the original Clicky by @FarzaTV, but for Linux. No subscriptions needed, just have a text-vision model setup on your consumer GPU #rtx3090. Tested with Qwen 3.6-35B and Gemma 4-31B. You can find it here: https://t.co/vVB0eKoUaU
i dream of running sota level llms on a single 3090. and i can see it coming, day by day.
i've been running gpus since 2017 and the 3090 joined my stack when it launched in 2020 and never left. the same card that made me a living back then is making me one through research now.
people ask what to do when one card from 5 years ago can do this much.
i might not know everything about local llms but after thousands of sessions i'm certain the 3090 is deeply untapped. especially when you stop relying on pre-made frameworks and go kernel-level. almost nobody has explored that path well. that's what keeps me awake. i'm not done with this card.