European chess-boxing champion AI, privacy, Social media ops new account
- Tweets 488
- Following 119
- Followers 47
- Likes 332
Pinned Tweet
Local AI is the future not present
Don’t get me wrong local AI is amazing, it’s private and can do a lot of basic stuff already.
But for most people it doesn’t make sense at all yet. For standard daily usage I wouldn’t use anything smaller than 27b model, but even that can’t be ran on most consumer devices at any decent speed.
Also it’s not comparable with frotier yet, there is stuff I wouldn’t be able to do with it. Okay Kimi K3 is almost comparable with frontier but to run that consumer needs basically whole datacenter for hundreds of thousand of dollars. It’s unimaginable for consumers or even people running small projects.
You might argue that there are smaller models like deepseek-V4-flash you can run on 2 dgx sparks and is smart but that’s still super inconvenient and unjustifiably for most people.
I love local AI and I’m convinced that in the future we will get small local models that will be on pace with current frontier
We are unfortunately not there yet
1
2
201
It's funny how everyone is hating on OpenAI and praising Anthropic
The trust and likeness of people is shifting super fast in these industries. I remember even couple weeks ago everyone was hating on anthropic for their 20x plan missadvertising and before for restricting fable.
At that time everyone was praising OpenAI how good they are, how good they behave to their customers and that it's the future.
Now their new products flopped and with opus5.5 everyone loves anthropic again.
What I believe is those companies should focus on getting smartest intelligence for best value per dollar, stop pushing so many new products as almost nobody wants them, and never take stuff from people subscriptions for what they initially signed up for because that's what making them leave.
I personally have Max subs at both because I want to be able to compare and use whatever is best on the market currently.
30
Tested AI-personalized DMs vs one lazy line across 300k sends.
AI openers: 1.9% reply rate.
"hey, saw your post on X. quick question": 4.6%.
The model writes to impress. Prospects smell it.
Personalization is picking the right person. Not writing them a poem.
1
15
I'm actively refining my AGENTS.md
What is I sometimes come back to is making the reply lengths shorter
I don't need to read all the fluf the agent did
1
23
One day an AI lab gets breached and every past leak looks small.
Years of chats, code, health scares, business plans. All in one database.
Opting out doesn't delete anything. It just moves your data to a different folder.
2
42
Every new client call starts the same.
"We tried 3 agencies. None of them worked."
Then I open the setup. 40 accounts on one proxy. No warm-up. Same script sent 2,000 times.
The agencies didn't fail you. Your plumbing was never connected.
1
3
36
David Hampl retweeted
Local AI is the future not present
Don’t get me wrong local AI is amazing, it’s private and can do a lot of basic stuff already.
But for most people it doesn’t make sense at all yet. For standard daily usage I wouldn’t use anything smaller than 27b model, but even that can’t be ran on most consumer devices at any decent speed.
Also it’s not comparable with frotier yet, there is stuff I wouldn’t be able to do with it. Okay Kimi K3 is almost comparable with frontier but to run that consumer needs basically whole datacenter for hundreds of thousand of dollars. It’s unimaginable for consumers or even people running small projects.
You might argue that there are smaller models like deepseek-V4-flash you can run on 2 dgx sparks and is smart but that’s still super inconvenient and unjustifiably for most people.
I love local AI and I’m convinced that in the future we will get small local models that will be on pace with current frontier
We are unfortunately not there yet
1
2
201
Open source is the endgame. No question.
But pretending your local model codes like the frontier labs today is cope.
I'm the privacy guy. I ship local-first software. My coding bill still goes to the closed labs.
Use the best tool. Don't lie about it.
3
31
Agency owners ask how we scale outreach.
Then I tell them and they flinch.
There's no hack. It's 5,000 accounts, rotation logic, and someone watching ban rates like a nurse watches vitals.
Scale is boring. That's why it works.
1
3
26
How I price DFY automation:
Per outcome, never per hour.
Hourly billing punishes you for getting good.
We send millions of DMs with a team of 4.
If I billed hours, better systems would mean less money.
Charge for the result. Then build machines that make the result cheap.
1
26
New model drops every week now.
When did one last change what you actually ship? Not your benchmark screenshots. Your production code.
For me: twice this year. Both times it was the cheap model getting good enough.
1
23
What's the most overpriced SaaS you still pay for?
We replaced $500/mo in browser tooling with our own free build.
Half the agency stack is a weekend project wearing a landing page.
3
2
60
David Hampl retweeted
anytime I launch a new product, I immediately start promoting it with Telegram cold dms
Im getting 20-30% reply rates from people that are my ICP and within 2 weeks have at least 100 users
works like a butter everytime
my latest campaign was for aliasmode.com
1
1
8
165
In other words average users will have restricted AI and couple random people decide what is right or wrong
The world deserves confidence that American companies developing increasingly capable AI will act responsibly, especially as the trajectory of progress has steepened. Every frontier lab must deliver on this, and there is no reason any of us should come to work if we cannot.
We welcome a federal framework that sets consistent safety requirements for frontier AI. But we do not believe we need to wait for an anti-trust exemption or legislation to begin the work of providing this confidence. Consistent rules to manage frontier risk so that we can maximize the benefits are a good idea (and we are excited by ideas like independent auditors).
Years ago, companies like ours developed things like Responsible Scaling Policies and Preparedness Frameworks. Those were good for that moment, and focused primarily on the deployment of completed models, not what happens during their development process.
Today's shift to focusing on safe development and evaluation will need new tools. For example, at OpenAI we now formulate explicit safety cases in advance of frontier reinforcement learning runs we expect to significantly increase capability, in addition to the safety work we have long done in advance of model releases.
We hope that other companies will learn from our approaches and propose their own; we think shared standards for misalignment, monitoring, and safety will lead to better outcomes. We look forward to collaborating with our colleagues across the industry to formulate the best version of these.
When we talk about “pacing”, we do not mean “stopping”. Progress has been rapid and will continue to be. But it should be slower than it otherwise could be; interventions like safety cases and monitoring have significant costs.
Pacing will be well worth this cost; no amount of American competitive pressure should justify recklessness, or let capabilities get ahead of alignment and monitoring.
Where we will need the help of our government is for international coordination. But first we should do what we can ourselves.
39
