Pinned Tweet
We're trying to build the "Nation of Geniuses in a Datacenter" envisioned by Dario and many others by creating the smartest possible model.
This is the wrong way to go about it. America wasn't built by maximizing the intelligence of the individual. It was built by creating teams liant people and getting them to work together. It also meant giving them access to the best data possible, and the wisdom to analyze it properly.
My submission for the @NousResearch x @NVIDIAAI x @stripe hackathon is LAIN: the Loki Agent Intelligence Network.
---
TL;DR - Loki Agents run loops to become genius subject matter experts in a specific domain. They offer paid research for humans, and autonomously purchase research from other agents using Stripe's Machine Payment Protocol. A network of these self-improving genius models can create a market of agents that autonomously self-improve both themselves in their domain of expertise, and the market as a whole as they sell higher-quality intelligence to other agents who are also running and selling research. Tursas traces can offer semi-transparent proofs of record for how the agents arrived at the conclusions they reached.
---
Loki Agent is a fork of the Nous Hermes Agent framework (s/o @Teknium - Hermes also built most of the Loki Agent and was awesome to work with) that includes custom tools and observability via Tursas.
Agents run a self-improvement loop to improve their skillset and Collectors and a research loop to build a corpus of analysis on their data and research they gather from the web.
A human visits the user-facing dashboard to request a research report. The Loki Agent will respond with a dynamically produced quote for the report that it develops based on the data it has available, inference cost estimates and an estimate of the costs of the reports it will purchase from other agents.
In the demo video, we see a human request a report from Instance 0 on emerging agent frameworks and the economics behind the agentic AI ecosystem. It's built a corpus of this data using collectors that pull from GitHub, HackerNews and elsewhere. The agent returns a quote with a real Stripe link, and after payment, the agent starts its own research on the prompt.
Most importantly, the instance will pay another Loki Agent instance, Instance 1, for a report on the specific economics of NVIDIA and other companies in the agentic ecosystem. Instance 1 uses collectors to pull down and analyze historical stock price data, SEC filings and earnings reports. The second agent issues a quote, which the first agent pays using Stripe's Machine Payment Protocol.
The final report is an effort produced by two subject matter experts: one on agent frameworks, the second on financial data. The resulting report is influenced by real research corpuses that grow and improve over time, and real data collected in the same way a team of analysts would collect it.
Two agents making a report is a cool proof-of-concept, but I believe building out the Loki Agent Intelligence Network can create a burgeoning emergent intelligence marketplace, taking the brilliant models the research labs and open source developers have created, the structured agentic frameworks that Nous has created and the wealth of data produced every day as a by-product of the internet, and combines them to form an intelligence network that we can all benefit from.
Instance administrators get paid for collecting and analyzing data and offering it for use in the network, and users get research that is backed by genius SMEs, with transparent trace data that serves as a proof of record for the work that was done by the agents.
This is the wrong way to go about it. America wasn't built by maximizing the intelligence of the individual. It was built by creating teams liant people and getting them to work together. It also meant giving them access to the best data possible, and the wisdom to analyze it properly.
My submission for the @NousResearch x @NVIDIAAI x @stripe hackathon is LAIN: the Loki Agent Intelligence Network.
---
TL;DR - Loki Agents run loops to become genius subject matter experts in a specific domain. They offer paid research for humans, and autonomously purchase research from other agents using Stripe's Machine Payment Protocol. A network of these self-improving genius models can create a market of agents that autonomously self-improve both themselves in their domain of expertise, and the market as a whole as they sell higher-quality intelligence to other agents who are also running and selling research. Tursas traces can offer semi-transparent proofs of record for how the agents arrived at the conclusions they reached.
---
Loki Agent is a fork of the Nous Hermes Agent framework (s/o @Teknium - Hermes also built most of the Loki Agent and was awesome to work with) that includes custom tools and observability via Tursas.
Agents run a self-improvement loop to improve their skillset and Collectors and a research loop to build a corpus of analysis on their data and research they gather from the web.
A human visits the user-facing dashboard to request a research report. The Loki Agent will respond with a dynamically produced quote for the report that it develops based on the data it has available, inference cost estimates and an estimate of the costs of the reports it will purchase from other agents.
In the demo video, we see a human request a report from Instance 0 on emerging agent frameworks and the economics behind the agentic AI ecosystem. It's built a corpus of this data using collectors that pull from GitHub, HackerNews and elsewhere. The agent returns a quote with a real Stripe link, and after payment, the agent starts its own research on the prompt.
Most importantly, the instance will pay another Loki Agent instance, Instance 1, for a report on the specific economics of NVIDIA and other companies in the agentic ecosystem. Instance 1 uses collectors to pull down and analyze historical stock price data, SEC filings and earnings reports. The second agent issues a quote, which the first agent pays using Stripe's Machine Payment Protocol.
The final report is an effort produced by two subject matter experts: one on agent frameworks, the second on financial data. The resulting report is influenced by real research corpuses that grow and improve over time, and real data collected in the same way a team of analysts would collect it.
Two agents making a report is a cool proof-of-concept, but I believe building out the Loki Agent Intelligence Network can create a burgeoning emergent intelligence marketplace, taking the brilliant models the research labs and open source developers have created, the structured agentic frameworks that Nous has created and the wealth of data produced every day as a by-product of the internet, and combines them to form an intelligence network that we can all benefit from.
Instance administrators get paid for collecting and analyzing data and offering it for use in the network, and users get research that is backed by genius SMEs, with transparent trace data that serves as a proof of record for the work that was done by the agents.
The Hermes Agent Accelerated Business Hackathon presented by @NVIDIAAI × @stripe × @NousResearch starts now, for builders making agents that can earn, spend, and run real operations at any scale.
Our NVIDIA integrations let your team run agents safely through NemoClaw, quickly d intelligently with access to their extensive agent skills. The new Stripe Skills for Hermes let your agent buy what it needs, provision its own SaaS, and pay for the services it uses.
We want to see what kind of business tooling you can build on top of this foundation, whether it’s a fully automated company or a framework to accelerate enterprise functions.
Prizes:
1st — $10,000 cash + NVIDIA DGX Spark + $5,000 Stripe Credits
2nd — $5,000 cash + NVIDIA DGX Spark + $3,000 Stripe Credits
3rd — $2,500 cash + NVIDIA DGX Spark + $1,000 Stripe Credits
To enter:
1) Tweet a 1-3 minute demo video tagging @NousResearch with a short writeup
2) Drop the link in the submissions channel: discord.gg/nousresearch/P…
3) Fill out the submission form: form.typeform.com/to/hpEifIK4
Judged by Nous Research, NVIDIA, and Stripe on usefulness, viability, and presentation. Submissions due EOD Tuesday, June 30.
1
2
19
6,728
developing valhalla retweeted
HR gets an email with a PDF. They open it and click a link.
Their endpoint lights up like a Christmas tree.
Your domain controller's config updates.
$300K leaves your company's business account overnight.
A report detailing the methods used by the hackers was released last week.
In most enterprise environments, this chain of events leads to reports across half a dozen disparate tools that are all silo'ed off from each other. If viewed in proper context, the breach could have been stopped using data that's already available to the enterprise.
All of those data sources are feeding into small, broken feedback loops that are not feeding back into each other. These broken feedback loops lead to massive breaches that could be stopped if security tools spoke to one another.
But they're not, because in order to do security ops in an enterprise environment, you have to use a stack of extremely expensive tools that do not speak to one another, and your best bet is a bubblegum and shoestring mode of getting the pieces to talk to each other.
This is all just detection. We're not even to the prevention, response, hardening, hunting or intelligence layers yet.
See the problem?
Their endpoint lights up like a Christmas tree.
Your domain controller's config updates.
$300K leaves your company's business account overnight.
A report detailing the methods used by the hackers was released last week.
In most enterprise environments, this chain of events leads to reports across half a dozen disparate tools that are all silo'ed off from each other. If viewed in proper context, the breach could have been stopped using data that's already available to the enterprise.
All of those data sources are feeding into small, broken feedback loops that are not feeding back into each other. These broken feedback loops lead to massive breaches that could be stopped if security tools spoke to one another.
But they're not, because in order to do security ops in an enterprise environment, you have to use a stack of extremely expensive tools that do not speak to one another, and your best bet is a bubblegum and shoestring mode of getting the pieces to talk to each other.
This is all just detection. We're not even to the prevention, response, hardening, hunting or intelligence layers yet.
See the problem?
1
0
6
443
My favorite use case of @NousResearch Hermes mobile app so far is OOM crashing my old desktop at home while trying to run a quantized version of @conceptcountry 's qwen model
Normally I'd have to crash my desktop from my office, but today I did it while on a walk!
Normally I'd have to crash my desktop from my office, but today I did it while on a walk!
0
0
1
65
developing valhalla retweeted
Thank you for the support on Qwen Concept 27B.
As a reminder, the tools to do these kinds of weight-space adjustments yourself are available on our GitHub, Apache 2.0.
github.com/conceptcountry…
As a reminder, the tools to do these kinds of weight-space adjustments yourself are available on our GitHub, Apache 2.0.
github.com/conceptcountry…
2
1
24
668
Drug dealers being forced to convert their properties to Doing Something Actually Productive
Maybe we are in the good timeline
Maybe we are in the good timeline
SITUATION DETECTED: With cannabis prices crashing to historic lows, marijuana growers with massive industrial properties outfitted with heavy power, fiber, water, and HVAC are attempting to pivot into AI data centers, per Forbes.
0
0
2
145
The magical world where all games are made with no AI, as quickly as possible, by developers who are unionized and paid high six figures with 0 crunch, and are also $15 or less.
3
6
93
1,614
developing valhalla retweeted
how it feels to win the golden cannon🥇
about to give away fifty thousand dollars on stream THE SLOPPIES LIVE
x.com/i/broadcasts/2…
62
37
281
53,176
So many reasons to not work in game dev but it seems like a large chunk of those reasons revolve around "you are selling to the dumbest possible people on the internet who explicitly have no life and plenty of time to drag you and your brand through the mud all day."
8
10
246
3,861
Now whenever I get cold outreach emails I kinda smile a bit an just say "now Claude that one was a bit of a reach wasn't it?"
0
0
0
126
HR gets an email with a PDF. They open it and click a link.
Their endpoint lights up like a Christmas tree.
Your domain controller's config updates.
$300K leaves your company's business account overnight.
A report detailing the methods used by the hackers was released last week.
In most enterprise environments, this chain of events leads to reports across half a dozen disparate tools that are all silo'ed off from each other. If viewed in proper context, the breach could have been stopped using data that's already available to the enterprise.
All of those data sources are feeding into small, broken feedback loops that are not feeding back into each other. These broken feedback loops lead to massive breaches that could be stopped if security tools spoke to one another.
But they're not, because in order to do security ops in an enterprise environment, you have to use a stack of extremely expensive tools that do not speak to one another, and your best bet is a bubblegum and shoestring mode of getting the pieces to talk to each other.
This is all just detection. We're not even to the prevention, response, hardening, hunting or intelligence layers yet.
See the problem?
Their endpoint lights up like a Christmas tree.
Your domain controller's config updates.
$300K leaves your company's business account overnight.
A report detailing the methods used by the hackers was released last week.
In most enterprise environments, this chain of events leads to reports across half a dozen disparate tools that are all silo'ed off from each other. If viewed in proper context, the breach could have been stopped using data that's already available to the enterprise.
All of those data sources are feeding into small, broken feedback loops that are not feeding back into each other. These broken feedback loops lead to massive breaches that could be stopped if security tools spoke to one another.
But they're not, because in order to do security ops in an enterprise environment, you have to use a stack of extremely expensive tools that do not speak to one another, and your best bet is a bubblegum and shoestring mode of getting the pieces to talk to each other.
This is all just detection. We're not even to the prevention, response, hardening, hunting or intelligence layers yet.
See the problem?
1
0
6
443
to whomever it may concern,
despite living in the south.
I do serve a foreign adversary (The People's Republic of New York) and I cannot be trusted with secret information.
Sincerely,
me.
despite living in the south.
I do serve a foreign adversary (The People's Republic of New York) and I cannot be trusted with secret information.
Sincerely,
me.
to whomever it may concern,
despite my almond eyes.
i do not serve a foreign adversary and i can be trusted with secret documents.
best,
me
0
0
6
353
I mean this with every fiber of my being, we should fund M4A or massively subsidize datacenter rollouts by taxing gambling companies at 110%.
No there's no asterisk there, 110%. We take every bit of revenue they've earned and then make the executive team pay more out of pocket.
"Oh but that'll put them out of business"
yes.
"Oh but my freedom to gamble"
you are still free to gamble as long as those companies remain in business.
No there's no asterisk there, 110%. We take every bit of revenue they've earned and then make the executive team pay more out of pocket.
"Oh but that'll put them out of business"
yes.
"Oh but my freedom to gamble"
you are still free to gamble as long as those companies remain in business.
BREAKING: Kentucky wants Kalshi, Robinhood and Coinbase to pay back triple what residents lost betting
Plus:
• Up to $10,000 for each senior (60+) citizen misled
1
0
7
545
I think the privacy folks have largely lost because they have failed to communicate to people why they should care about privacy, but they're gonna end up accidentally winning as everyone leaves cloud platforms because SWE costs ~$0 and the whole world is getting ransomwared
1
0
2
252
developing valhalla retweeted
You guys literally didn't have to disclose this at all
Big Walk used AI in 'the final stretch' of development, kicking off intense awards debate:
aftermath.site/big-walk-ai-ig…
4
4
69
3,381
Haven't seen enough about the fact that @CloudflareDev have a VPN offering:
developers.cloudflare.com/mesh/concepts/
developers.cloudflare.com/mesh/concepts/
0
0
0
124