GraphicalChops retweeted
After 10 years of being in crypto and trading, I need to address what's currently happening in our space.
This isn't just another market cycle - it's something far more concerning.
The Reality Check:
Meme coin trading has become a 24/7 sprint where:
• You're considered "late" after just 15 minutes
• Holding for 6-12 hours is seen as "long-term"
• Every new token launch feels like a trap
Let's be honest about what we're seeing:
1. The Speed
These aren't market cycles anymore - they're market seizures. When you need to be glued to your screen every second just to avoid being "late," something's fundamentally broken.
2. The Scams
The number of rugs, honeypots, and outright frauds is staggering. It's not just about "doing your research" anymore when you have minutes to make a decision.
3. The Burnout
I've traded through bull markets, bear markets, and everything in between. But this? This is different. The constant rotation, the 24/7 FOMO, the endless parade of "next big things" it's exhausting even the most seasoned folks.
4. The Community Impact
We're burning out builders who are chasing a dragon that doesn't exist. Real projects can't compete with pump-and-dumps masquerading as "communities."
Remember Warren Buffett's wisdom: "Only when the tide goes out do you discover who's been swimming naked."
Well, the tide is starting to shift, and we're seeing a lot of naked swimmers.
The Hard Truth:
• This isn't sustainable
• Good projects are being overshadowed
• Experienced traders are burning out
• The community is losing faith
It's going to get worse before it gets better.
But we need to start having honest conversations about where we're heading.
What We Need:
• Return to fundamentals
• Focus on sustainable growth
• Support for real builders
• Community-driven accountability
For those feeling the weight of this market:
You're not alone. It's okay to step back. It's okay to miss a pump. Your mental health matters more than any potential gains.
The market will be here tomorrow. Make sure you are too.
GraphicalChops retweeted
1) Fun fact: there are ~280 payment companies today building on cryptorails (!)
I've written a piece explaining the current financial system, major use cases for cryptorails, adoption challenges they're facing, and what the future might look like.
paragraph.xyz/@archetype/cry…
GraphicalChops retweeted
"Verticalized agents" is a bit of a meme among startups right now—but rightly so: Over the last year, we indexed too far into AI being a Swiss army knife—jack of all trades and master of none.
You simply cannot go wrong building something obsessively focused around a problem. And this is not simply because you will deliver value more precisely—but because the world is still behind in realizing what AI can do and your marketing message needs to spoonfed them: "DM and pitch creators on TikTok", "create HTML animations with speech", etc.
AI or not, my rule of thumb has always been that Version 1.0 of any product should be so verticalized that:
1) when a target user lands on the page, their only reaction is "Well, shit. This was made for me. I better sign up."
2) when you tell your friends & family what you're working on, it is so narrowly focused that you feel embarrassed
GraphicalChops retweeted
The biggest opportunities for AI Agents will be in expanding the market for software by automating work that most companies don’t do that well or never even did in the first place.
This will take on a number of forms but there are at least 3 big categories that highlight how much larger these markets will be as AI fully rolls out.
1. Verticals. Every industry has dozens or hundreds of workflows that are highly domain specific and where both the customer and the company have just expect it’s supposed to take forever or be painful. Processing a loan or an insurance claim, reviewing due diligence documents, generating a grant proposal, transcribing a patient visit, and thousands more. These are the processes full of tons of drudgery and have never been automated well in the past. When AI automates them, these will represent net new areas of software spend.
2. Startups and small businesses. Most startups and SMBs are strapped for resources. This means they don’t tend to have a lot of the expertise that larger companies have in legal, finance, HR, sales, or any number of topics. AI provides an instant ability to augment their own team as they scale up. This makes it faster and easier to start a company and will drive significant AI usage.
3. Teams in larger companies. We tend to think about the “enterprise” market monolithically but even within enterprises there’s an entire universe of funding allocation decisions that mean some teams have all the resources and others don’t. AI Agents let teams that that were previously strapped for resource finally solve their problems better. This could be building custom software for some bespoke workflow or automating a contracting management process in a non-core area.
The opportunity is to find industries or market segments that have always had painful tasks that never get done efficiently and automate them. This will produce markets far larger than we realize.
GraphicalChops retweeted
It’s not in your head—but it’s also not in your genes.
After 40 years of research, Dr. Gabor Maté found that childhood trauma is the hidden root of all disease, addiction, and suffering.
Here are 10 mind-blowing lessons he teaches about healing trauma and reclaiming your life: 🧵
GraphicalChops retweeted
Y Combinator JUST announced what startups they want to fund next in 2025. And it's mostly AI that replaces $100k/year job functions.
My notes below in case it's helpful to you:
GraphicalChops retweeted
As we are learning DeepSeek is one of the most sophisticated psyops of all time.
Here's how it went down:
1) Release the model open source.
2) Include highly detailed papers for all other people to replicate your work.
3) Create a novel SOTA RL algo that uses less memory and bandwidth than PPO and release it absolutely free so anyone can copy it for their own work.
4) Make sure it's super easy to fine tune so that others can easily SFT out any local law abiding answers about Tiannamen in about half a day for $50 in compute.
5) Cackle while saying muahhahaha you can only run it on our servers. Oh wait actually you can run it on any hardware you want, fine tune it, replicate our entire method on your own models, which lifts up the entire research community.
Truly one of the most dastardly and sophisticated plans to come out of China in a long time!
GraphicalChops retweeted
The burst of DeepSeek V3 has attracted attention from the whole AI community to large-scale MoE models. Concurrently, we have been building Qwen2.5-Max, a large MoE LLM pretrained on massive data and post-trained with curated SFT and RLHF recipes. It achieves competitive performance against the top-tier models, and outcompetes DeepSeek V3 in benchmarks like Arena Hard, LiveBench, LiveCodeBench, GPQA-Diamond.
📖 Blog: qwenlm.github.io/blog/qwen2.…
💬 Qwen Chat: chat.qwenlm.ai (choose Qwen2.5-Max as the model)
⚙️ API: alibabacloud.com/help/en/mod… （check the code snippet in the blog）
💻 HF Demo: huggingface.co/spaces/Qwen/Q…
In the future, we not only continue the scaling in pretraining, but also invest in the scaling in RL. We hope that Qwen is able to explore the unknown in the near future! 🔥
💗 Thank you for your support during the past year. See you next year!
GraphicalChops retweeted
Here is the most cost effective way to deploy (the real) R1 in the cloud - use llama.cpp-powered inference endpoints @huggingface
The @UnslothAI quantizations fit neatly in 4x L40S
(max 8192 ctx for now)
Use the following link:
endpoints.huggingface.co/new…
GraphicalChops retweeted
Just sat down to read the DeepSeek-R1 paper. We're entering an era where compute isn't primarily for training. It's for creating better data. I expect to see the money & compute spent on data processing (generation / annotation / curation) grow to match and exceed the money & compute spent on pre-training.
People have talked about pre-training plateauing because we're "running out of data" on the internet to scrape. While that may be the case, capability improvements are going to continue full steam ahead. The improvements in intelligence are going to come not from putting in more data (scraped from the internet) but rather from putting in more compute (to generate higher-quality data). Intuitively, this feels similar to me to how people learn. You don't learn just by ingesting lots of tokens. In many cases, you learn by thinking more (I am referring to training time, but thinking more also applies at inference time).
There are many creative ways to put in more compute to get better data, and this problem will be an important research area for a number of years.
- In this paper, they train two models. Why two models? The first one (a reasoning model trained via RL) is used to generate data to train the second. This works by using the first model to generate reasoning traces and then selectively keeping only the high quality outputs (quality is judged by simply checking the results). This approach of "checking the results" works well for domains like math and coding where you can easily check the results.
- In drug development, it is super common to put compute into generating better data in two phases. In the first phase, a generative model (e.g., for protein sequences) generates a massive number of candidate drugs. In the second phase, scoring or filtering is done with a slew of predictive models which may predict structure, toxicity, solubility, binding affinity, etc. After all this work is done, you may end up with 100 data points.
- In physical domains (e.g., climate applications), expensive but accurate physics simulators exist (these simulations are run on super computers for long periods of time to simulate the physics of the atmosphere or some other system). All of that data can be used to train models, which is showing a ton of promise.
The question of "how to put more compute into generating better data" is central to progress in AI right now.
GraphicalChops retweeted
Jevons paradox is real.
After we released stable diffusion as an open source, smaller, faster, better version of the closed image models out there it completely used up all GPU capacity on clouds.
You couldn’t get NVIDIA chips for inference.
Market for thinking AI way larger.
Jevons paradox strikes again! As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can't get enough of. en.m.wikipedia.org/wiki/Jevo…
GraphicalChops retweeted
TLDR on what’s happening in ai for anyone out of the loop:
>sam altman shows up
>hello we are openai (is closed source btw) and we will be building ai
>sam: we need BILLIONS in funding!
>everyone: ok here you go sam
>sam: wait really?
>everyone: yea..? didn’t you say you need billions of dollars in funding?
>sam: billions? sorry I misspoke. I meant TRILLIONS!
>everyone: that’s a lot but ok
>sam: ai is the new industrial revolution! COMPUTE is the new OIL!
>nvidia becomes the most valuable company on earth. every single employee is a MILLIONAIRE now, including the interns
>openai keeps jacking up their prices and says “we are actually LOSING money because compute costs so much! we are bumping up the pricing for our gpt-abc123 model to $10,000 per prompt”
>some random chinese quants: we ain’t paying that much let’s just build our own model
>chinese quants literally build AGI (DeepSeek) as their side project with $70 in funding and a bag of chips, coworking out of a sweatshop
>american ai companies: these chinese guys are capping this is a ploy by the CCP to destroy the american economy!
>deepseek releases papers and benchmarks proving their model is better than openai’s
>american ai companies in total meltdown
The bubble has popped.
If you’re in ai, it’s OVER. Leave while you can! All the speculative institutional capital is going to rotate back to the only speculative technology that has stood the test of time: crypto.
GraphicalChops retweeted
If you're using Deepseek r1,
Try this *PPFO* Prompting Framework
We achieved o1 level response quality in @CodeGuidedev and reduced API cost by 50%
Tips:
- Use XML tags instead of Markdown
- No bullet points in purpose and output sections
- Provide as much detail as possible
GraphicalChops retweeted
Massive Opportunity:
If you know how to "fine-tune" AI models - it's your time to make $.
Just fine tune Deepseek r1 for companies!
This model is the first open-source model that is consistently good on local servers.
Charge $20k to $50k for each fine-tuning job.
GraphicalChops retweeted
Thirdweb just launched an SDK for on-chain AI agents, called "Nebula".
It says the model is trained on "billions of transactions, and millions of contracts".
Nebula can:
◆ Read on-chain data in real-time
◆ Execute transactions from a wallet
◆ Plug into Eliza/Langchain & more
GraphicalChops retweeted
A website that aggregates all AI agent tools in one place
• Frameworks
• Launchpads
• Tools
• Tutorials
• Hackathons
• Research
• Infrastructure
• People
• Apps
• Agents
aiagenttoolkit.xyz/
GraphicalChops retweeted
BREAKING: Nvidia, $NVDA, announces Project Digits personal computer at $3000, that is approximately 1,000 times more powerful than the average laptop.
The device is powered by an Nvidia GB10 Grace Blackwell Superchip, which houses separate, linked components on a single chip to reduce the time it takes to move data between them.
The superchip features an Nvidia Blackwell graphics card and an Nvidia Grace processor, packaged with 128 gigabytes of memory and 4 terabytes of SSD storage.
