@AlexFinn the wild part is 1% to 2% sounds small but thats literally double the demand overnight
we are NOT ready for that supply shock
English
Prompt To Point
172 posts
Prompt To Point
@PromptToPoint
Turning generic AI content into sharper ideas, better prompts, and stronger visuals. Less slop. More signal.
London Присоединился Temmuz 2024
7 Подписки4 Подписчики
5 months ago I spent $30,000 on 3 Mac Studios, 2 Mac Minis, and a DGX Spark
I went all in on local LLMs and encouraged others to do the same
I warned prices would explode
I was called crazy, a hype beast, dangerous, and that I had no idea what I was talking about
Since then:
• Mac Studios above 96gb have become unavailable
• Memory prices have 4x’d
• Other hardware prices have 10x’d
Now those same AI influencers who destroyed me are spending 5 to 6 figures on hardware publicly
GLM 5.2 dropped and it’s Opus level. I’m running it on 1 of my 3 Mac Studios 512gbs. The same ones I was called an idiot and hype beast for buying. The same ones that are reselling for triple the price used.
The insane part is this is just the beginning
Intelligence will be integrated into every device you own, including devices that aren’t even publicly available yet like humanoid robots
All of these new devices will require GPUs, memory, storage, and more components
Components that have already 10x’d in price
That’s not even counting all the people that will start vibe coding when Codex and Claude Code become more mainstream
Right now less than 1% of the world is even taking advantage of those tools
Imagine what happens when it reaches 2%
The local revolution is here. Hardware is the bottleneck
Act accordingly
English
@exploraX_ imagine being a vc-backed SaaS founder and then THIS list drops
the whole business model just got open sourced lmao 🔥
English
10 GITHUB REPOS THAT SHOULD BE ILLEGAL TO HAVE.
all free. all open-source. bookmark this for later.
1️⃣ recordly — the free screen studio.
open-source screen studio. auto-zoom, smooth cursor, webcam overlay, styled backgrounds, polished demos without an editor. (AGPL-3.0)
🔗: github.com/webadderallorg…
2️⃣ stirling-pdf — your entire pdf toolkit
self-hosted. merge, split, sign, redact, OCR, convert, compress, 50+ tools, runs locally, nothing leaves your machine. (MIT)
🔗: github.com/Stirling-Tools…
3️⃣ photogimp — turns GIMP into photoshop. photoshop shortcuts, layout, and splash screen patched right on top of GIMP 3+.
🔗:github.com/Diolinux/Photo…
4️⃣ open notebook — self-hosted notebooklm.
drop in pdfs, urls, youtube links — chat with them, summarize, even generate podcasts. bring your own model (18+ providers). (MIT)
🔗: github.com/lfnovo/open-no…
5️⃣odysseus — pewdiepie's self-hosted AI workspace.
chat, agents, deep research, docs, email, memory — local-first, your hardware, your data. (MIT)
🔗: github.com/pewdiepie-arch…
6️⃣freedomain — free domain names for everyone.
register a domain, point it at cloudflare or any DNS, ship your site without paying for the name. (AGPL-3.0)
🔗: github.com/DigitalPlatDev…
7️⃣ hyperframes — write HTML, render video.
heygen's engine that turns html/css + animations into deterministic mp4s. built for AI agents. (Apache-2.0)
🔗:github.com/heygen-com/hyp…
8️⃣ web-to-app — turn any website into an android app, on-device.
configurable webview, apk signing, even node/php/python runtimes — no remote build.
🔗: github.com/shiaho777/web-…
9️⃣ reclip — self-hosted video + audio downloader.
paste a link from youtube, tiktok, x, ig — 1000+ sites — grab it as mp4 or mp3. powered by yt-dlp.
🔗: github.com/averygan/reclip
🔟 excalidraw — the infinite whiteboard that replaces miro, figjam and lucidchart. hand-drawn diagrams, wireframes, real-time collab, end-to-end encrypted. 120k+ stars for a reason. (MIT)
🔗: github.com/excalidraw/exc…
most people pay monthly for tools that already exist for free. you don't have to.
m0h@exploraX_
10 free github repos that can replace major SaaS with subscriptions. all free. open-sourced. some are MIT licensed. — 1️⃣ openscreen — replaces screen studio ($29/mo) - a clean macOS/windows/linux screen recorder for polished demos. - blur, cursor highlighting, annotations, export to mp4 or gif at any aspect ratio. - doesn't try to clone every feature, just nails the basics for quick walkthroughs you'd post on X. — 2️⃣ voicebox — replaces elevenlabs ($22/mo) + wisprflow ($15/mo) - local-first AI voice studio. - clone voices from 3 seconds of audio, generate speech across 7 TTS engines in 23 languages, - dictate into any text field with a global hotkey. - nothing leaves your machine. - runs on apple silicon, cuda, rocm. — 3️⃣ openshorts — replaces opus clip ($19/mo) + submagic ($16/mo) - free AI video platform. - clip generator turns long youtube videos into 9:16 shorts with auto-subtitles and face tracking (runs on free gemini + elevenlabs tiers). - also includes AI UGC video generation with actors — that part is pay-per-use via fal. ai (~$0.65-2 per video). docker self-host. — 4️⃣ freellmapi — replaces chatgpt pro + claude pro ($20/mo each) - stacks 14 free AI provider tiers (google, groq, cerebras, openrouter, github models + 9 more) behind one openai-compatible endpoint. ~800M tokens/month. - smart router with failover, sticky sessions, encrypted key storage. ships with a dashboard. — 5️⃣ playwright-mcp — replaces browserbase ($39/mo) + browser use ($25/mo) - microsoft's official MCP server that gives any AI agent full browser control. - uses accessibility trees, not screenshots — deterministic and token-efficient. - works with claude code, cursor, windsurf, codex out of the box. — 6️⃣ vibe-trading — replaces tradingview premium ($60/mo) - natural-language finance research agent. - 7 backtest engines across stocks, crypto, futures, forex. - 75 specialist skills (factor analysis, options strategy, ML strategy). - 29 multi-agent swarm presets. - 21 of 22 MCP tools work with zero API keys. — 7️⃣ CalCom — replaces calendly ($12/mo) + savvycal ($12/mo) - the open-source scheduling infrastructure. - one-on-ones, group events, round-robin, team booking, - payment collection (stripe), routing forms, workflows. - integrates with google/outlook/apple calendar, zoom, meet, teams. - self-host in 10 minutes with docker. 40k stars. — 8️⃣ whisper — replaces otter ($17/mo) - openAI's open-source speech-to-text model. - transcribe audio in 99 languages, translate to english, generate timestamps. - runs locally on cpu or gpu. - the actual model behind most "AI transcription" SaaS tools you're paying for. — 9️⃣ postiz — replaces buffer ($15/mo) - AI-powered social media scheduler. - cross-post to X, linkedin, instagram, tiktok, threads, bluesky, mastodon, youtube, pinterest. - AI captions and hashtags. - analytics dashboard. team workspaces. 31k stars and rising. — 🔟 vaultwarden — replaces 1password ($8/mo) - unofficial bitwarden-compatible server written in rust. - works with every official bitwarden client (mobile, desktop, browser). - unlimited users, unlimited vaults, full enterprise feature set. - runs on a $5 VPS or your home server. — disclaimer: open-source ≠ 1:1 replacement. you'll trade polish for ownership, hand-holding for control, and a credit card for a github version. for builders, prototypers, and indie hackers — that's the whole point. for everyone else, the paid tools still have their place. bookmark this. share with one friend bleeding subscription fees. ~m0h
English
@EcZachly the math aint mathing the same way it used to
whole playbook got patched
English
The path of “grind leetcode to get a big tech job and work for 15 years and retire at 35” is dying.
Meta moved half their engineering team to training AI to replace them
Anthropic is a trillion dollar company with only a few thousand employees 1/5
English
@sudoingX the 'rent only the ceiling' mental model is genuinely underrated
most ppl optimizing for access when they should be optimizing for ownership 💯
English
cancel your chatgpt subscription for a month. buy a single used 3090, call it a grand. run qwen 3.6 27b dense on it and let it grind on your actual work, the code, the drafts, the boring research.
here's what happens. you go the whole month and barely hit a wall. the few times you do, you clock that THAT's the 10% you actually needed the frontier for, and the other 90% a card sitting in your room handled just fine.
most people pay every month for capability they touch a handful of times. own the 90%, rent the rest only when you hit the wall. trust me anon, you won't look at that subscription the same again.
English
@gokulr so the actual meta is using them to check each other's work
two AIs keeping each other honest is lowkey the most 2026 workflow ive seen 🔥
English
After using Claude Code (Opus 4.7/4.8) and Codex (GPT-5.5) incessantly for the past several weeks, my verdict is that (like every human) their greatest strength is also their greatest weakness.
tl;dr Use Claude Code + Opus 4.7/8 for brainstorming and planning. Use Codex + GPT 5.5 for execution and building. Use both to adversarially review each other's plans / design docs.
CC is really creative and a great brainstorming partner. However, this creativity makes it hallucinate when executing.
Codex is an incredible, focused, fast executor and builder. However, this makes it poor at generating new, creative options.
(I have friends at both OpenAI and Anthropic who agree with the above and use the "other" lab's product for precisely the use cases that their product is not good at).
English
@ihtesham2005 the scariest part is he probably genuinely believes the framework is airtight
no villain arc, just a closed loop that keeps confirming itself
English
Chamath fed Dario Amodei's own essays into Claude and asked for a psychological profile. What came back should be required reading for every investor in frontier AI.
The model identified a pattern.
Dario distrusts other labs. He distrusts authoritarian states. He distrusts markets to distribute the gains fairly. He distrusts institutions to move fast enough. And after Mythos, he distrusts the government to wield power transparently. That is a very long list of untrustworthy actors.
The list of trustworthy ones is conspicuously short. And it has a suspicious tendency to resolve toward people who reason the way he does, operating under rules he helped design.
Claude named it precisely. Not megalomania. Epistemic exceptionalism.
The quiet, defensible conviction that disagreement is always downstream of error. That when your safety framework requires someone to hold the keys and your analysis keeps concluding every other key holder cannot be trusted, you have built a machine that outputs the same answer no matter what you feed it.
The tell was a single word.
When the Mythos situation collapsed, Anthropic called it a misunderstanding. That word choice under pressure assumes that if everyone simply understood correctly, they would agree with him.
Sacks put it simply on the pod. They believe AI is super dangerous and only they are virtuous enough to control it.
That is not a safety framework.
That is a monopoly with a philosophy attached.
WATCH THE FULL PODCAST ON @theallinpod
English
@_MaxBlade open source just walked into claude's house and started rearranging furniture
the gap is closing SO fast omg 🔥
English
@givros this is the kind of honesty that actually makes you trust someones content long term
the 40 min grind is the skill, not the first prompt 💯
English
GPT-5.6 Pro is insane.
It didn’t just beat Claude 5, Fable and Mythos.
It crushed them.
One prompt, just 5 minutes. And voila:
A full low-poly Three.js island game with water, beach bar, boat, shark, fish, whales, sports area, NPCs and Ibiza vibes
This is getting ridiculous.
That is what I would write if I was not honest with you.
That is what I would post if I only wanted a viral clickbait tweet.
And this is exactly why you should always be careful with what you read on social media, especially on X.
There is no GPT-5.6 yet.
This prototype was made with GPT-5.5, in 6 iterations, after more than 40 minutes of back-and-forth with Codex.
Yes, it started with one prompt, and I will share that prompt below.
But the final result came from many corrections, adjustments, fixes, visual tweaks, Three.js issues, camera changes, object placement improvements, and a lot of manual direction.
That is the real workflow.
I share build in public content here to show what you can actually do with AI, but also to show that there is no magic dust.
AI is powerful.
Very powerful.
But the difference between a cool demo and a real usable prototype is still iteration, feedback, taste, debugging, and direction.
Yes, I am also waiting for GPT-5.6.
It will probably arrive next week, and when it does, we will test it here properly, in public, with real prompts, real iterations, and real results.
English
@AlexFinn bro it deployed to vercel AND picked the right domain BY ITSELF
we are not ready for how fast this goes 🔥
English
Codex is the best agent harness out there right now
If you are using it just for coding, you're doing it all wrong
Literally ALL work you do on your computer you should try to do through Codex first
I had to launch a new newsletter landing page today. I told Codex what I needed to do. Had no expectations it would work
It autonomously wrote the code, uploaded it to my github, went to my Vercel account, started a new project, connected the repo to vercel, then chose the correct domain name from my list in Vercel and connected it.
5 minutes after sending the prompt I went from nothing to a fully live working email collection landing page
I'd do these steps:
1. Write down every task you do on your computer
2. Before doing the task, start a new chat in Codex and tell it about the task
3. Ask how it would do it
4. Make sure browser use and computer use are enabled
5. Run every task and see how much it can complete
6. For repeating tasks, schedule them in an automation
Almost all your knowledge work on your computer will become automated. I'm very blown away
English
@0x_kaize free tier is sitting on a benchmark KILLER rn
the $20/mo crowd is gonna feel this one late 😭
English
GLM 5.2 just beat GPT-5.5 on code - and you can run it for FREE
the model that everyone will be paying $20/mo for in 6 months is currently sitting on the FREE tier
what GLM 5.2 actually did:
- SWE-bench Pro: 62.1 (GPT-5.5: 58.6)
- Terminal-Bench 2.1: 81.0 (4 points above Claude Opus 4.8)
- FrontierSWE: 74.4% - basically tied with Opus 4.8
- 1M context, open weights, ~1/6 the cost of GPT-5.5
two ways to actually use it:
[ FREE PATH ]
1/ go to zenmux(.)ai
2/ sign up with gmail
3/ grab an API key
4/ paste the base URL into Claude Code / Cursor / Cline / Hermes etc..
[ PRODUCTION PATH ]
1/ go to creao(.)ai
2/ open the model picker
3/ select GLM 5.2
4/ get 1M context + long-horizon agentic coding
5/ pay $4.40/M output vs Opus 4.8's $25/M - same intelligence, 1/6 the cost
6/ no rate limits when the free tier dries up
FREE for testing & @CreaoAI for shipping
most devs are still paying $20/mo for GPT-5.5 plans and getting less capability and don't want to try something new that actually works
stop thinking and start building already
English
@LuizaJarovsky 42 AGs AND a 25% extinction risk sitting in the same conversation is wild
how is that not the loudest alarm bell going off rn 😭
English
42 state attorneys general are investigating OpenAI over alleged harmful practices.
Read again:
42 state attorneys general are investigating OpenAI over alleged harmful practices.
The tide has totally changed. Even the authorities have had enough.
AI should exist solely for the purpose of societal good.
If, as Dario Amodei says, there is a 25% chance of AI-driven extinction, then:
No evidence of direct societal benefit, no AI.
Authorities should not accept anything other than that.
English
@DanielTNiles micron up 297% YTD and expectations are STILL high
that MU earnings reaction is gonna go crazy either way 🔥
English
Last wk, the Wednesday FOMC mtg led to a sell-off in both bonds and stocks but Thursday saw a rebound leading to a solid wk w/ S&P +0.9% led by Semi Index +7% & oil down 9% to $78. The yld curve though did flatten last wk with 2y +10 bps while 30y -7 bps.
As I previewed last week going into the FOMC mtg, I am more concerned about bonds than I am about stocks given the lessons from the 2013 “taper tantrum.” 9 of 18 Fed members are now looking for at least one rate hike in 2026.
On a positive note, the US reached a Memorandum of Understanding with Iran. Though the situation may be getting messy over the weekend yet again, as I have said numerous times, US political parties wanting to get re-elected during mid-terms is a powerful motivator to ultimately end this.
WTI rose from $67 on Feb 27th prior to the war to a high of $113 on April 7th and ended the week at $78. Interestingly while oil is still up 16% from 2/27, the S&P Energy subsector is down 5%. When the spread gets this wide versus the underlying commodity, the risk vs reward gets compelling for a trade. There are also likely to be a few more twist and turns before the MOU leads to a final agreement so it also acts as a hedge if the war escalates yet again before it gets better.
On the negative side, the rise in another commodity, semiconductors, is starting to lead to price hikes as talked about by the CEO of $AAPL last week. While the buyers of iPhones are in a higher income bracket and therefore less price sensitive, on the margin prices going up in the consumer electronics industry in general is likely to hurt consumer demand.
In addition, I am starting to increasingly worry about the implications of token maxing which started in late March turning to token minimization as companies try to get their AI costs under control. While maxing should have helped Q2 results at the big hyperscalers, minimization could cause Q3 guides to be potentially less robust. But keep in mind, this is only occurring at the top companies on the leading edge of AI while 95% are still trying to figure out how to use Agentic AI. Leading companies are also trying to get token costs under control by using orchestration layers to route to cheaper open-source LLMs when appropriate
But if the bullish argument on why the AI bubble is not going to burst any time soon is that there is not enough compute to satisfy all the current demand, then what is the logic of cutting prices as supposedly OpenAI is thinking about. Even if it is to gain share at enterprises against Anthropic, they should not be able to service any incremental demand generated if compute is the bottleneck.
Memory stocks have rightly been on fire this year given that is one of the big bottlenecks for AI compute. Therefore it will be interesting what $MU has to say when they report earnings on Wednesday and how the market reacts. With the semiconductor index up 102% year-to-date and Micron up a blistering 297% the expectations are clearly high. It has only risen once in reaction the next day to the past six earnings reports. As an example, $JBL which reported a solid beat and raise quarter last Wednesday morning due to hyperscaler momentum saw their stock open 10% higher, trade to up 14% intra-day, but ultimately close a touch lower.
So in summary I remain bullish longer-term based on S&P earnings growing over 25% in 2026 and the ramp of Agentic AI. But in the near-term, I am increasingly expecting a “speed bump” given the risks discussed above.
Happy Father’s Day and all the best in the week ahead.
English
@nabeelqu the pangram check idea is so obvious it hurts
just pick a lane and enforce it, this ambiguity is embarrassing 💯
English
*Another* apparently AI-generated story wins a literary prize, this time judged by a panel including the novelist Ruth Ozeki.
Literary prizes need to start including Pangram checks in their process, or else change the rules to make AI writing ok. It’s very simple!
English
@JeremyNguyenPhD @emily_beam @girayaslim academics finally showing the rest of us how to actually use these tools properly
slides from a real webinar?? thats a gold mine 🔥
English
Claude Code and Codex for Research
@emily_beam and @girayaslim from University of Vermont share slides from last week's webinar "Agentic Tools for Research"
Links in the replies below:
English
@kimmonismus anthropic playing long game while openai just wants the ticker symbol first
different games entirely
English
@james406 LMAOOO they multiplied
worst break-in ever or best?? 😭
English
seriously be careful out there everyone
i had 2 Microsoft Copilot licenses in my car, and someone broke in and left 4 more
English
@arpit_bhayani generating a fresh uuid INSIDE the retry loop is wild to me
thats literally just giving up on the whole idea
English
AI agents will retry. They will always retry. We have to be ready for it. Given how long-running agentic loops are, network drops, timeouts, and rate limits kicking in are super common.
When that happens, the agent often does not know if the original tool call succeeded. So it calls again. Thus, if that tool charges a card, sends an email, or creates a record, a retry means it happens twice.
The fix is to make every mutating tool call idempotent. The second run of the tool should be a no-op that returns the original result. The standard way to do this is an idempotency key.
The caller generates a unique key per logical operation (a UUID tied to the user's intent, not the HTTP request). We store the result keyed by that value. If a request with the same key arrives again, the server simply returns the stored response.
Idempotency key generation logic completely depends on what task you are doing, but a few patterns cover most cases.
In some cases, the client generates a stable operation ID. In others, a deterministic key can be derived from the business operation being performed.
For example, in a one-shot user action (place an order, send a payment), hash the meaningful inputs together: user ID, action type, and the request payload.
A common mistake is generating the key from something that changes on every attempt, like a timestamp or a fresh UUID with no link back to the original call. That defeats the purpose. The key has to be deterministic from the standpoint of "what is the agent trying to accomplish," not "when did this packet leave the machine."
The key has to live above the retry loop, not inside it. Generate it once, before the first attempt, and pass that same key into every retry of that operation. The network can fail ten times. The key does not change.
Idempotency is not a nice-to-have for agents; it is a must-have when it comes to building reliable agents.
Hope this helps.
English
@burkov the verification signal doing more heavy lifting than anyone wants to admit
true/false is secretly the most powerful label in ML rn 🔥
English
Someone asked how a Chinese company managed to catch up to Codex and Claude Code in coding. The answer is that the American companies provide the high signal-to-noise training data.
The way it works is as follows (all is scripted, no human in the loop):
1. You take a large enough base model and finetune it using a combination of reinforcement learning and supervised finetuning.
2. To get training examples, you ask some LLM to introduce a subtle bug into an existing codebase and provide a test script that would only return True when the bug is fixed.
3. You use Codex/Claude Code to fix the bug. When they do that, you record all LLM inputs and outputs.
4. You use these inputs/outputs for supervised finetuning of your model.
5. You use the test script result (True/False) for reinforcement learning.
Supervised finetuning trains the conversational part of the problem solving. Reinforcement learning trains the actual problem-solving part.
This way, you don't train a weaker student. Verifiable results (True/False) combined with a solution space exploration technique is what eventually makes the student stronger than the teacher.
Coding LLMs is a solved problem and everyone will end up having the same solution.
English
@0xFrogify bro fluffy the ai bunny is eating harder than most startups 💀
why am i still overthinking everything 🔥
English
She’s making $10,000 a month with a cartoon bunny.
Not a real character.
Not a big team.
Just an AI bunny named Fluffy that she created in 7 minutes.
1/4 - She takes viral kids videos
2/4 - Throws the link into ChatGPT
3/4 - Gets a prompt, and generates the whole thing inside Picsart Flow.
4/4 - Some of her videos are pulling hundreds of millions of views.
YT pays from $1,000 to $10,000+ for 1 million views on long format vids
While most people are still trying to come up with original ideas and burning out…
She’s copying what’s already working, turning it into AI content, and letting it print.
This is one of the fastest and dumbest ways to make money on YouTube right now.
Most people will watch this and think it’s too good to be true.
The ones who actually try it are going to be in a completely different position in a few months.
Watch exactly how she’s doing it 👇
Follow me if you want to see EXACT step-by-step guide. I will post iton my X shortly, notifs on!
Frogify@0xFrogify
English
@zarazhangrui codex just sitting there being a full upgrade and people still sleeping on it
chatbot era is cooked fr
English
People should just use Codex as ChatGPT
Since I started using Codex/Claude Code daily, I rarely open ChatGPT these days (unless it’s a quick online search type of task, in which case I use Gemini)
The coding agent’s result is usually strictly better than a chatbot’s
English