Manu | 🥥

12 AI tools that help you work less and scale faster No fluff. No hype. Just tools that get sh*t done 👇 Let’s break them down:

Meta just showed something more important than a new model they’re making intelligence economically scalable here’s what actually matters: -> same capability with ~10x less compute -> RL that scales predictably (not experimentally) -> reasoning that’s controlled at inference time but the real shift is this: models are no longer just optimizing for intelligence they’re optimizing for when and how much to think multi-agent systems are just the mechanism the real innovation is managing reasoning under cost constraints that’s what will define real-world AI systems

@AIatMeta This is the real signal here: not better models better coordination multi-agent test-time reasoning is probably the most important part of this thread we’re moving from “thinking longer” to “thinking together” huge shift

To build personal superintelligence, our model’s capabilities should scale predictably and efficiently. Below, we share how we study and track Muse Spark’s scaling properties along three axes: pretraining, reinforcement learning, and test-time reasoning. 🧵👇 Let’s start with pretraining. Over the last 9 months, we rebuilt our pretraining stack with improvements to model architecture, optimization, and data curation, enabling us to increase the capability we can extract from every unit of compute. To rigorously evaluate our new recipe, we fit a scaling law to a series of small models and compare the training FLOPs required to hit a specific level of performance. The results: we can reach the same capabilities with over an order of magnitude less compute than our previous model, Llama 4 Maverick, making Muse Spark significantly more efficient than the leading base models available for comparison.

@AIatMeta Which niche is Meta trying to target with this new system? I think this whole application to the Ads system is very powerful, especially on platforms like Instagram.

@4evernoeyy It had to be said, and it was said haha 100% agree with you..

@ManuAF6 Good question… I think they’re building for institutions first, not individual builders like us.

Claude just crossed a line most people still don’t understand This is not about “AI helping you code”... Claude Mythos is reportedly finding exploits in major websites in ~72% of trials Not assisting -> Finding vulnerabilities We’re no longer measuring models by: - reasoning - coding ability - benchmarks We’re measuring them by: their ability to break real systems And that changes everything. Because finding vulnerabilities requires: - understanding complex systems - identifying edge cases - simulating adversarial behavior This is closer to offensive security than AI assistance. If this holds outside controlled environments: Software security becomes asymmetrical. One model can scan thousands of attack surfaces continuously without fatigue Now the real question it's: Who should be allowed to use them?

This is bigger than “Meta launched another model” Muse Spark suggests Meta is trying to win AI from a different angle: - Native multimodality - Parallel reasoning / multi-agent orchestration tool use - Distribution across Meta’s consumer ecosystem If this gets deeply integrated into Instagram, WhatsApp, Facebook, Messenger and glasses, the real moat won’t just be model quality It’ll be context, distribution, and everyday usage 🔥

@4evernoeyy 100%! But my question remains... "What target audience is Anthropic building for?"

@ManuAF6 Feels like the real advantage isn’t Mythos alone, but the system wrapped around it.

@josephweb3 Boss!!

i'm successful on my trades lately i am not even good at technical analysis > focus on people's sentiments > this war is the main thing that moves the whole market (commodities, stocks, and cryptos). > i think timezone has a huge role too i was awake during the attack by the USA and Israel. i was awake when POTUS dropped a deadline. and awake when POTUS wants a 2-week ceasefire.🫡 which are the outlets i am following? @sentdefender @MarioNawfal @spectatorindex and a few more (for the fastest news) for fast confirmation @Cointelegraph @WatcherGuru @AshCrypto and a few more but beware, a lot of the content they post are noise too which doesn't move the market. eg. Israel continues airstrike against Hezbollah, and Iran continues barrage on UAE, Qatar, etc, these aren't moving the markets, unlike when the US or Trump did something crazy. these are just some of the trades i took profit lately. while my long position after the ceasefire confirmed are still on.

This is the first time a model isn’t released because of what it can actually do That’s the signal: From Intelligence to Control 2 years ago, we worried about: wrong answers hallucinations Now? real-world impact – finding exploits – chaining actions – running autonomously

@josephweb3 Fully agree. When AI makes vuln hunting cheap, defense must evolve beyond controls and regulations hahaha

@ManuAF6 100% when vuln hunting gets cheap, defense has to change too.

@4evernoeyy Exactly @4evernoeyy : the offense-defense gap flips when defenders patch faster Mythos finds zero-days, but Glasswing closes the discovery-disclosure loop-turning scary AI into boring, reliable security

@ManuAF6 This could make systems stronger long-term if defenders adopt it faster than attackers.

@DataChaz FAAAAACTs

accurate

@sherifgjini Ability to delegate

If you had to choose one founder skill for the next 12 months: - coding - distribution - sales - prompting

@feifei_qiu It should have been a long time ago... but we're not ready to dedicate time to learning... Now everyone wants to learn because they've been given the shortcut... Typical of humans

Coding soon will become one of the core literacy skills just like reading, writing, and speaking.

@JonhernandezIA It's just a matter of marketing and targeting the audience.. Give them what they want, show them what they need

It's just me feeling OpenAI is becoming the Ai you can afford while anthropic what you actually want? This is gona be true for so many people, and this is gona make the world so much worst. The underclass begins

@NFTFlow_ @BawsaXBT That happens when you read that X has now boosted interaction... but... there's a keypoint in the algorithm that they're unaware of. It's not just about the responses... but about what you're responding to hahaha

@BawsaXBT think some people are heavily relying on engagement groups + just spamming replies to try get replies back on their own sh*t 👀

if someone’s post gets 100+ likes within the hour of it going out, is that real motion or a sign of botting? cause i be doing my DD and it seems like half of these people got bots interacting with their posts. activating my inspector gadget 🧐

@boxmining From Premium, to freemium, to tokenium, to free? Let's go hahah

GPT stepping up. If you can't compete on max intelligence, just give users free stuff.

Makes sense my friend.. But once this capability exists, control doesn’t scale It spreads, gets replicated, and eventually moves outside regulated environments In my opinion.. The real shift is that the cost of finding vulnerabilities is collapsing So the system itself needs to adapt, not just access

100% they can definitely break many system and exploit vulnerabilities. companies who will be using such models must have some rigorous background check, history check, and even regulation on using them. anthropic itself must put some limitation with the companies who will use their model. eg. they need to submit the software they will be scanning and reason why... eg. paid to scan and test security, etc...