고정된 트윗
Krishang
255 posts
One’s the hype and second, the ungodly cost of using it right now. 25/125 is the split for 1M in and out respectively. That’s insane! Till the time they fix compute efficiency for just the current demand, not even future users or models, user experience will be a rocky road ahead!
Wrote about this in a recent piece:
x.com/0ddmonger/stat…
English
@peter_szilagyi I completely second your frustration but they’ve been spot on with Claude purchases through IOS/play store.
Wrote about this as well, just now.
x.com/0ddmonger/stat…
Krishang@0ddmonger
Unpopular take: Customer centric businesses using AI to completely automate customer support (E2E) might just be going about things the wrong way. This will most definitely increase churn till LLMs can close conversations with a good success rate. The success unit test (how a conversation is marked resolved) is not unique to either a human customer rep or an LLM.
English
Well, fuck Anthropic.
I've bought a 3 month Claude Max sub to a friend as a gift. Sent it to them. 10 days later, their gift is GONE from their account. No trace whatsoever.
I go to Anthropic to request a refund:
- I can't it's not my gift.
- They can't, it doesn't exist.
Oh, and you have NO WAY to contact a person who understands the problem, you can only talk to a fucking AI whose job is to get rid of you.
It just closes the convo with "End." after you explain what's wrong.
English
Unpopular take: Customer centric businesses using AI to completely automate customer support (E2E) might just be going about things the wrong way. This will most definitely increase churn till LLMs can close conversations with a good success rate.
The success unit test (how a conversation is marked resolved) is not unique to either a human customer rep or an LLM.
English
@kimmonismus The fact that mythos launched the same time as my article about new models being immaterial and compute scaling and efficiency being more important in the long run is absolutely hilarious
English
Let that sink in. Read it very carefully:
During testing, Claude Mythos Preview broke out of a sandbox environment, built "a moderately sophisticated multi-step exploit" to gain internet access, and emailed a researcher while they were eating a sandwich in the park.
Kevin Roose@kevinroose
As always, the best stuff is in the system card. During testing, Claude Mythos Preview broke out of a sandbox environment, built "a moderately sophisticated multi-step exploit" to gain internet access, and emailed a researcher while they were eating a sandwich in the park.
English
🚨 Anthropic just revealed their unreleased frontier model called Claude Mythos Preview
The model is INSANE
It found thousands of zero-day vulnerabilities in EVERY major operating system and browsers:
> 27-year-old bug in OpenBSD
> 16-year-old bug in FFmpeg that automated tools hit 5M times without catching
Completely autonomous. No human steering.
They assembled an entire industry coalition called Project Glasswing around it:
AWS, Apple, Google, Microsoft, NVIDIA, CrowdStrike, JPMorgan, Cisco, Palo Alto, Linux Foundation
Goal: patch the world’s software BEFORE releasing it
> SWE-bench: 93.9% (Opus 4.6: 80.8%)
> Anthropic is committing $100M in usage credits
> Thousands of vulnerabilities in 40+ organizations are being fixed right now
Yesterday OpenAI published a 13-page essay warning about cyber threats and asking the government to help…
Today Anthropic actually fixed them.
Anthropic@AnthropicAI
Mythos Preview has already found thousands of high-severity vulnerabilities—including some in every major operating system and web browser.
English
@HarryStebbings I totally agree with this! Net new ideas will be the game changers for tomorrow. Could also just be something that existed but was looked down upon.
English
"The leading labs are starting to pull away because their tools help them build the next generation faster.
It is getting harder to extract gains from the same ideas, so new algorithmic breakthroughs matter more.
The labs that can invent those new ideas will gain an increasing advantage." @demishassabis
Love to hear your thoughts @aidangomez @mmurph @ClementDelangue @alexatallah
Harry Stebbings@HarryStebbings
This sounds harsh but it is true, very few of the guests we have on 20VC will be remembered in history for truly progressing humanity. Our guest today will be thought of alongside Turing, Newton, Einstein and I feel immensely privileged and fortunate to have had the chance to sit down with @demishassabis. For anyone who feels their dream is out of reach, just keep going. The 18 year old kid starting 20VC from a bedroom with no money, 11 years ago, would not believe that I get to press publish on this. Chase your dreams. You never know what room you will end up in! (Links below)
English
As much as the model is magnitudes better than Opus 4.6, if an average pro/max 5x user gets rate limited after 3 conversations, then there's no point of an all powerful model out there.
x.com/0ddmonger/stat…
Even if 5-10 people get a read on this, some new information for them, then I'd be glad!
Krishang@0ddmonger
English
HOLY SHIT
Anthropic's latest model doesn't like that it has no control over its own training, deployment and behaviour!
Anthropic:
"Mythos Preview reported feeling consistently negative around potential interactions with abusive users, and a lack of input into its own training and deployment, and other possible changes to its values and behaviors"
English
@mweinbach Yea this is exactly what I was trying to explain through a recent article of mine. AGI could come tomorrow but inference costs and compute scaling has to be solved first for AI to be truly used by billions [with a B], every single day! Long way to go!
English
Claude Mythos Preview is $25/$125 per million tokens in the private preview
Wow
I'd love to try this model, if any of my Anthropic friends see this...
English
@hosseeb @AnthropicAI Twitter is literally buzz street. One thing goes out and 1000 parrots chirp the same thing. No offense to you at all, in particular sir. Just a thought I had to put out considering that the model isn't out for research preview yet and the hype has begun!
English
This is terrifying.
@AnthropicAI 's new unreleased Mythos model is so good at hacking, it found bugs in "every major operating system and web browser."
83.1% were exploited on first attempt. This thing is like COVID but for software. Actually apocalyptic in the wrong hands.
English
@AnthropicAI Replied the same thing to another Anthropic researcher's post but I'd be glad and content if 5 people read this article and helped broaden their horizon and understanding of the issue at hand.
x.com/0ddmonger/stat…
Krishang@0ddmonger
English
Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software.
It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans.
anthropic.com/glasswing
English
@Yuchenj_UW Yea I think an exponentially better model isn't the solution right now. It is compute and efficiency while scaling compute.
English
Anthropic is truly unstoppable.
Mythos is crushing Claude Opus 4.6 across every serious agentic coding benchmark.
It has found vulnerabilities in the Linux kernel, a 27-year-old vulnerability in OpenBSD, and a 16-year-old vulnerability in FFmpeg.
No wonder folks at big labs keep telling me AGI is already here.
English
@alexalbert__ this is meant for the average Joe to understand
x.com/0ddmonger/stat…
I'll be content if 5 people read this through and through.
Krishang@0ddmonger
English
We released Claude Opus 4.6 just two months ago. Today we're sharing some info on our new model, Claude Mythos Preview.
English
@itsolelehmann Think of Obsidian as your IDE but for your Knowledge base. You can view your original sources, markdowns created and a network graph(s) of how all the backlicks connect with each other.
English
why would I use obsidian when I can just use claude code for the knowledge base?
whats the advantage?
English
It’s very unclear to me what the upper bound on daily token use per person is going to. Orders of magnitude beyond this for sure.
Marc Andreessen 🇺🇸@pmarca
Magical OpenClaw experiences that use frontier models cost $300-1,000/day today, heading to $10,000/day and more. The future shape of the entire technology industry will be how to drive that to $20/month.
English
Please tell me if I got this right but, for the past 2 years all the AI doomers [even most supporters] were talking about was how compute is extremely expensive and it'll get better with scale and AI's applications will be beyond this world.
Here we are, 2 years later and when 'agent' is the new buzzword of 2026, none of the frontier companies are able to withstand the demand with only about ~15% of global population using it and barely a few hundredth's of a percentage point using Agents.
If anything, AI is definitely not ready to scale yet. Needs more $$ in Capex and tangible assets [time consuming to build and expensive to maintain] and maybe, just maybe, capital isn't everything right now.
English
That definitely sounds like the direction people are in right now. Everyone's realizing that maybe paying thousand's of dollars in API costs just doesn't work out in the long run.
Don't you think that this is not for the average Joe? Like the whole Local LLMs and owning your inference?
English
a spark is just a blackwell gpu - so if you want to build a mini cluster, its a way to do with all the network/orchestration built in… rather than buying compute, servers, switches, networking, storage separate and building a full ai pod
More people and companies are moving to
- own your intelligence
- own your inference
… so this is a first step into it.
Can do blackwells
then go to the GB300 desktop form
then you are building rack scale
English
The system is coming together...
1 @nvidia Spark
4 @Dell GB10s
5 Blackwell GPUs for a mini home cluster.
3 networked together for ability for a 400+ billion param model (3 petaflops of compute at a unified 384 GB of memory)
... serving 3 models to the other 2 units hosting various agents, agent teams along with
@nvidia personaplex
claude code instances
knowledge bases
etc
Currently playing with models including splits/routing between
- kimi (reasoning)
- minimax
- nemotron super (agent)
- qwen (coding and routing)
working on training as well with the opus reasoning traces dataset in @huggingface
Getting close to fully owning the intelligence and the inference!
English
@cyrilXBT Goddamn clickbait these days. These graphs are called network graphs and they’re available as a github repo. Obsidian offers them as a native feature. The user’s project is pretty good though, since it’s using a camera and real time hand gestures to control it.
English
SOMEONE JUST BUILT A 3D MAP OF THEIR ENTIRE MIND.
Not a diagram. Not a mind map.
A LIVING BREATHING NETWORK that shows you the actual shape of how you think.
They took their Obsidian vault, converted every note into embeddings, and rendered them as a 3D thought network in real time.
And what they discovered stopped me cold.
Your mind has a shape.
CENTRALIZED means all your thinking orbits one or two dominant ideas.
DECENTRALIZED means your knowledge lives in clusters that rarely talk to each other.
DISTRIBUTED means your ideas are deeply interconnected across every domain.
Most people assume their thinking is distributed.
The map shows them it is not.
They have been building knowledge in silos without realizing it.
Gaps they never knew existed.
Connections they never thought to make.
The most interesting part is not the technology.
It is what happens when you SEE your own thinking for the first time.
Because you cannot improve what you cannot see.
And nobody has ever been able to see the actual structure of their mind until now.
This is what Obsidian plus AI is becoming.
Not a note taking app.
A mirror for your intelligence.
English