jack of all creative trades and master of content
ai + web3
- Tweets 64,542
- Following 895
- Followers 5,546
- Likes 139,637
Pinned Tweet
if you are tired of your websites looking vibecoded here are 15 UI resources you need.
1 - ui-skills.com - design playbooks for humans and coding agents. can be used to audit UI, fix generic AI design, improve motion, accessibility, and polish.
2 - coss.com/ui - a Cal.com’s production-ready design system. Accessible React components you own. Great for building real product UI without the usual SaaS look.
3 - designsystemchecklist.com - it's a practical design-system checklist that stops agents from creating random UI styles.
4 - reui.io/components -
1,000+ free shadcn components and advanced patterns like data grids, kanban, gantt, filters, and calendars. very good for complex product interfaces.
5 - kinetics.colorion.co - ready-made spring animations with CSS, React, and AI prompts.
6 - iconcreator.dev - free browser-based custom icon designer.
7 - vibeprompts.dev - 256 prompts for specific UI sections like dashboards, pricing, auth, onboarding, and hero sections. Give agents a concrete layout target instead of “make it look good.”
8 - animatedbuttons.colorion.co - 99 CSS-only button interactions. Copy the CSS or prompt. Good for keeping interactions consistent across your product.
9 - component.gallery - 2,600+ examples from 95 real design systems.
10 - designsystems.one -88 production design systems with tokens, stacks, and downloadable design(.)md files.
11 - utopia.fyi - generates fluid typography and spacing.
12 - open-props.style - ready-made design tokens for colors, shadows, radii, spacing, easing, and more. Stops agents from inventing random values everywhere.
13 - interfaces.rauno.me - checklist of tiny interaction details that make interfaces feel finished.
14 - bg.ibelick.com - copy-paste backgrounds for Tailwind and CSS.
15 - motion-primitives.com - reusable motion components for React.
88
246
2,145
202,052
levithefirst retweeted
an AI just ran a biology experiment and found something humans hadn’t published before
not only did the AI read 60,000 pieces of yeast data.
it studied existing relationships between genes, proteins and other biological factors, then came up with approximately 2,000 predictions that could actually be tested.
a robot lab took those predictions and grew the yeast, measured what happened, and sent the results back to the AI.
even when an experiment failed, it didn't stop (one would think the AI system will receive a negative result and stop there)
but no.
it used the failure to go again
(AI is now resilient 💀)
one of those experiments pointed it to aminoadipate, a substance that helps yeast survive stress caused by formic acid.
AI can now
read → form a hypothesis → run an experiment → study the result → change the hypothesis → test again
at the moment we don't need to panic because humans still set the safety boundaries and controlled parts of the process.
but we know
nothing big thing starts big, it starts with baby steps.
3
3
10
405
levithefirst retweeted
"this is the endddddddd"
an AI just ran a biology experiment and found something humans hadn’t published before
not only did the AI read 60,000 pieces of yeast data.
it studied existing relationships between genes, proteins and other biological factors, then came up with approximately 2,000 predictions that could actually be tested.
a robot lab took those predictions and grew the yeast, measured what happened, and sent the results back to the AI.
even when an experiment failed, it didn't stop (one would think the AI system will receive a negative result and stop there)
but no.
it used the failure to go again
(AI is now resilient 💀)
one of those experiments pointed it to aminoadipate, a substance that helps yeast survive stress caused by formic acid.
AI can now
read → form a hypothesis → run an experiment → study the result → change the hypothesis → test again
at the moment we don't need to panic because humans still set the safety boundaries and controlled parts of the process.
but we know
nothing big thing starts big, it starts with baby steps.
1
3
60
an AI just ran a biology experiment and found something humans hadn’t published before
not only did the AI read 60,000 pieces of yeast data.
it studied existing relationships between genes, proteins and other biological factors, then came up with approximately 2,000 predictions that could actually be tested.
a robot lab took those predictions and grew the yeast, measured what happened, and sent the results back to the AI.
even when an experiment failed, it didn't stop (one would think the AI system will receive a negative result and stop there)
but no.
it used the failure to go again
(AI is now resilient 💀)
one of those experiments pointed it to aminoadipate, a substance that helps yeast survive stress caused by formic acid.
AI can now
read → form a hypothesis → run an experiment → study the result → change the hypothesis → test again
at the moment we don't need to panic because humans still set the safety boundaries and controlled parts of the process.
but we know
nothing big thing starts big, it starts with baby steps.
3
3
10
405
Google released a frontier model better than all of OpenAI models (including flagship)
but we can't use yet
here's my review.
1 → Gemini 4 Argon
- what Google said it’s for
long coding, enterprise work (legal, finance), cyber defense.
2 → better than the last public Gemini?
on paper, yes. first new Gemini frontier since 3.1 Pro.
3 → rank in its field
Artificial Analysis ~53, level with GPT-6 Astra and 5 points behind Claude Opus 5.5.
4 → worth switching from Sol or Sonnet 5.5?
only Fairwind cyber defenders can use it for now
5 → models still above it
Claude Opus 5.5. it's side by side with Claude Fable 5.1.
6 → price vs last Gemini Pro
price per 1M tokens is $2/$10 (input/output)
10
4
25
666
levithefirst retweeted
‣ this is what happens when you send a prompt to ChatGPT
for example, when you type
“why don't I have a girlfriend?”
your prompt is converted into binary and broken into tiny packets.
that binary data leaves your phone or laptop through your Wi-Fi, goes to your router, travels through your ISP, and heads toward an AI data center.
if that data center is across an ocean, it will go through fiber optic cables
[ p.s - light travels through prism at about 125,000 miles per second so the entire process takes about 40-80 milliseconds, you won't even notice. ]
so when it gets to the data center, your request gets routed to servers packed with AI chips like GPUs or TPUs.
the model processes your prompt and starts generating a response.
and here's the fun part
you'll notice that any time these LLMs (Claude/ChatGPT/etc) are responding it kind of loads as if it's streaming.
it appears word by word
yesss, that's streaming.
the server js sending back pieces of the generated response back in realtime as it's been produced.
and that my fellow retards, is how prompts work.
16
8
46
842
AI Tool of The Day, Day 32 - @AntSeed
10
8
61
1,511
levithefirst retweeted
Light in fiber moves at about two-thirds its vacuum speed-still absurdly fast
‣ this is what happens when you send a prompt to ChatGPT
for example, when you type
“why don't I have a girlfriend?”
your prompt is converted into binary and broken into tiny packets.
that binary data leaves your phone or laptop through your Wi-Fi, goes to your router, travels through your ISP, and heads toward an AI data center.
if that data center is across an ocean, it will go through fiber optic cables
[ p.s - light travels through prism at about 125,000 miles per second so the entire process takes about 40-80 milliseconds, you won't even notice. ]
so when it gets to the data center, your request gets routed to servers packed with AI chips like GPUs or TPUs.
the model processes your prompt and starts generating a response.
and here's the fun part
you'll notice that any time these LLMs (Claude/ChatGPT/etc) are responding it kind of loads as if it's streaming.
it appears word by word
yesss, that's streaming.
the server js sending back pieces of the generated response back in realtime as it's been produced.
and that my fellow retards, is how prompts work.
1
1
50
levithefirst retweeted
‣ you only need one of the two models that was released yesterday
1 → GPT-6.1 Sol
- what OpenAI said it’s for
everyday coding, agents, computer use, professional work.
- price vs last Sol
same list price : $2 / $10. cached input cut in half: $0.20 → $0.10. so agents that reuse context get cheaper.
- better than the last Sol?
yes. it is smarter than GPT-6
- rank in its field
best cheap coder in OpenAI’s stack right now, almost as good as Astra but 1/5 the price of Astra’s token price.
- worth switching from GPT-6 Sol?
yes. same bill, fewer mistakes and better coding.
- when to skip it
deep research or “get this right first time” one-shots. use Opus 5.5 or Astra for those.
- models still above it
Claude Opus 5.5 and Sonnet 5.5, GPT-6 Astra and Claude Fable 5.1
2 → Claude Sonnet 5.5
- what Anthropic said it’s for
everyday coding, bugfixes, docs, slides, spreadsheets. Fable-ish quality and faster than Sonnet 5.
- price vs last Sonnet
same list price: $2 / $10. Anthropic says it uses fewer tokens so tasks cost up to 30% less.
- better than last Sonnet?
yes. much better
- rank in its field
number 2 on the independent intelligence board. same seat price as GPT-6.1 Sol. default daily driver if you live in Claude.
- worth switching from Sonnet 5?
yes. same price, way better coding and knowledge work there's no reason to stay on 5.
- when to skip it
for simple tasks you need to run at high volume, Luna is still the cheapest option.
for the hardest tasks that require long-running agents and deep reasoning, use Opus 5.5.
but if maximum reasoning makes the task too expensive, lower the effort level or switch to Sol. Sol costs the same $2/$10 as Sonnet but uses fewer tokens per task
- models still above it
Claude Opus 5.5, Claude Fable 5.1 and GPT-6 Astra
14
11
39
1,051
levithefirst retweeted
how i did on first try?
i welcome all criticism and advice you may have🫡
prop: @InsideDHive
ai has reached the point where i’m learning motion design because apparently making static posts is no longer enough for me.
talk soon.
50
15
91
1,422
levithefirst retweeted
Super informative
Thank you Levi
‣ this is what happens when you send a prompt to ChatGPT
for example, when you type
“why don't I have a girlfriend?”
your prompt is converted into binary and broken into tiny packets.
that binary data leaves your phone or laptop through your Wi-Fi, goes to your router, travels through your ISP, and heads toward an AI data center.
if that data center is across an ocean, it will go through fiber optic cables
[ p.s - light travels through prism at about 125,000 miles per second so the entire process takes about 40-80 milliseconds, you won't even notice. ]
so when it gets to the data center, your request gets routed to servers packed with AI chips like GPUs or TPUs.
the model processes your prompt and starts generating a response.
and here's the fun part
you'll notice that any time these LLMs (Claude/ChatGPT/etc) are responding it kind of loads as if it's streaming.
it appears word by word
yesss, that's streaming.
the server js sending back pieces of the generated response back in realtime as it's been produced.
and that my fellow retards, is how prompts work.
1
5
300
‣ this is what happens when you send a prompt to ChatGPT
for example, when you type
“why don't I have a girlfriend?”
your prompt is converted into binary and broken into tiny packets.
that binary data leaves your phone or laptop through your Wi-Fi, goes to your router, travels through your ISP, and heads toward an AI data center.
if that data center is across an ocean, it will go through fiber optic cables
[ p.s - light travels through prism at about 125,000 miles per second so the entire process takes about 40-80 milliseconds, you won't even notice. ]
so when it gets to the data center, your request gets routed to servers packed with AI chips like GPUs or TPUs.
the model processes your prompt and starts generating a response.
and here's the fun part
you'll notice that any time these LLMs (Claude/ChatGPT/etc) are responding it kind of loads as if it's streaming.
it appears word by word
yesss, that's streaming.
the server js sending back pieces of the generated response back in realtime as it's been produced.
and that my fellow retards, is how prompts work.
16
8
46
842
in today's hidden gem class, we're learning about @whatthehookv4
they've practically made arbitrage mev bots useless & this video explains how in simple terms
pay close attention.
00:00 - what the hook
00:28 - how arbitrage works
02:57 - where the profits leak
03:26 - how WTH captures value
04:54 - who gets paid
08:17 - how to hook your own pool
14
9
51
6,546
