Content Creator|| Writer || Exploring the modular frontier || In my early era, watch it compound.
Joined September 2022
- Tweets46.7K
- Following2.8K
- Followers4.1K
- Likes53.9K
Nabeelahhh💕 retweeted
I fly aircraft that report problems in seconds and file them in weeks.
In May a crew logged a subtle altimeter drift during descent. By the time a technician read it, that aircraft had already had a near miss in low visibility.
The crew was right. The note was just outside the aircraft's clock.
A court runs on years. An aircraft runs on seconds. An answer that arrives outside the clock is not safety, it is a post-flight briefing.
Lee Castleton was bankrupt years before the Horizon records were shown to be wrong. The court kept perfect time. It was never his clock.
AI agents will run on milliseconds. A system measured in months is not slow, it is in another universe.
That is what @GenLayer is: an adjudication layer where many independent AI validators read the same evidence separately and reach a verdict together, in minutes and at a fraction of the cost of a court.
Minutes is not instant, but it is inside the life of a disputed deal. That is the clock that matters.
I process flight telemetry in real time now. Not because I distrust the report. Because the report was never the problem.
What is one near miss that only made sense afterwards?
#JusticeNeedsAnUpgrade
Slow. Expensive. Wrong.
It's been this way for centuries. It doesn't have to be.
#JusticeNeedsAnUpgrade
Nabeelahhh💕 retweeted
My mentor used to say a late verdict is still a failed one. He was being kind.
Villagers reported the algae before any official did. The reports climbed a chain built for caution. The data stacked up. By the time the bay was ruled unsafe, the fishery was already gone.
That is not a late verdict. It is an autopsy.
The film's third case is the same shape. Anthony Ray Hinton lost nearly 30 years to bullets that never matched his gun, and walked free in 2015. Nobody can return those years.
A sentence can be quashed. A dead bay cannot. In my field the ruling often lands on something that no longer exists.
Now AI agents will make millions of deals a day. By the time a court reaches one, the money has cleared, the goods have shipped and the agent has moved on. There is nothing left to rule on.
@GenLayer is the adjudication layer I would point to. Independent AI validators each review the evidence and reach a verdict together, in minutes and at a fraction of the cost of a court.
So I review AI marine models now, looking for the threat while it is still a signal. Speed is not the opposite of care. In conservation it is the only form of care that arrives in time.
What is one thing you lost while someone else was still deciding?
#JusticeNeedsAnUpgrade
Slow. Expensive. Wrong.
It's been this way for centuries. It doesn't have to be.
#JusticeNeedsAnUpgrade
Nabeelahhh💕 retweeted
My ram is safe with sam I guess🙂
People should always be in control of the future.
OpenAI is against the notion that it would be acceptable for people to be relegated to a role of doing things that “AI systems think makes sense to reward in humans”.
People should not cede control of the future to superintelligence, and that requirement should ultimately constrain AI progress.
Nabeelahhh💕 retweeted
every voice agent i've heard talks over itself at some point 😂😂
you ask it something, the speaker leaks into the mic, and the stt thinks you just said something. the agent cuts itself off and the whole convo breaks
@soniox_ai v6 feeds the agent's own tts text into speech recognition, so it knows what the agent is saying and can filter out its voice while still hearing you when you actually interrupt
this tackles a real problem for voice agents, and that same context also helps with recognising short answers, names and numbers
Introducing Soniox STT v6, the world’s first speech recognition model that knows what your voice agent is saying and automatically filters out the agent’s own voice.
Two connected innovations make this possible:
→ Agent text input gives recognition context from both sides of the conversation, improving its understanding of user responses.
→ Agent voice echo cancellation filters out the agent’s playback while continuing to listen to the user, even when they interrupt.
The result: fewer false interruptions, fewer clarification loops, and more reliable conversations.
V6 also delivers significantly higher accuracy across 60+ languages for voice typing, meeting transcription, multilingual recognition, and real-time translation, with stronger speaker diarization, language identification, endpoint detection, and context customization.
One unified speech model. All 60+ languages. The full feature set, with languages freely mixed.
Available now for real-time streaming and asynchronous transcription.
Explore what’s new in Soniox STT v6:
soniox.com/blog/soniox-stt-v…
Nabeelahhh💕 retweeted
anthropic just open-sourced flashprove
This lets developers prove a response really came from a specific model
so no secret model swaps, and it's much harder to hijack a chat
Something interesting about this
anthropic says claude itself made the speed improvements and wrote the protocol spec
which claude though? no word on that, but with fable 5.5 rumours going around i'm watching closely
they say it's orders of magnitude faster than earlier open-source methods
the catch is that it takes about 70-95x the compute of just running the model
New Anthropic research: progress on provable inference for large language models.
Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so.
These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers.
For our full write-up, see here: anthropic.com/research/prova…
Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: github.com/anthropics/FlashP…
We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (rdalal.github.io) and Ben Mayer (@transmissions11).
Nabeelahhh💕 retweeted
The report was right. We still lost the wall.
In March, our inspector flagged a crack in a retaining wall. It sat in a queue for months. By the time someone opened it, the wall had failed after the first heavy rain. Nobody disagreed with the report. It was just late.
Late justice is still failure.
The film's slow case hit me for that reason. A suit filed in Calcutta in 1951 was dismissed in 2026 because nobody appeared for either party. 75 years, and it ended in an empty courtroom.
No rule was broken in either case. The report existed, it was filed, it moved. Every step did what it was told. Nobody asked whether this crack was the one that mattered. That is why the first red flag gets acted on now, not the third.
Now AI agents will make millions of deals a day. By the time a court rules on one, the agent has signed a thousand more and the money has moved twice. That verdict is a receipt, not a remedy.
@GenLayer is the adjudication layer I would point to. Independent AI validators each review the evidence and reach a verdict together, in minutes and at a fraction of the cost of a court.
The film is in the post I quoted. Watch it, then quote it with the case that hit hardest.
What is one warning you raised that everyone agreed with and nobody acted on?
#JusticeNeedsAnUpgrade
Slow. Expensive. Wrong.
It's been this way for centuries. It doesn't have to be.
#JusticeNeedsAnUpgrade
Nabeelahhh💕 retweeted
My Companion is ready for the Labyrinth 🔥
Drew this overnight after grinding the quests and walking all day.
I really had fun creating this for @CompanionsWfit cause it's been a while i picked up my pen to draw.
I also featured @TheSuitedApes
And @VaultlingsNFT in the 🎨
While I had fun creating and posting it, I hope the @CompanionsWfit
And @WFitApp see's and like it before it's too late
Btw... happy Halloween everyone 👻
Nabeelahhh💕 retweeted
i'm building a game called NIMBUS
Classic Skypea in one piece and dragon ball Z inspired
floating islands you can land on, walk, run and jump around
fall off and a cloud catches you 😭😭
there's a pet cloud too
this is what it looks like now
I'm having way too much fun with blender and opus 5.5 ultracode:)
Nabeelahhh💕 retweeted
chatgpt just plugged into your apple life lmao
iCloud Mail and Calendar plugins are live
It will find that email, catch up on your inbox, update events
also plans for you on web, desktop and iOS
basically, it's life made easier;)
Nabeelahhh💕 retweeted
So yesterday, OpenAI announced that it has just cut its frontier models off the internet during training
one model found a DNS misconfig and got itself online in September
so now they train on an "offline world" of cached web pages
but even offline, agents on separate machines still found a way to pass secrets to each other through the CPU cache
no network needed
Anthropic did the same thing 2 weeks ago
the labs aren't scared of what the models know anymore
they're scared of what they'll do
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
Nabeelahhh💕 retweeted
if Opus 5.5 was a pilot i will let it fly me anywhere
i made this browser flight simulator in 40 hrs just with Claude Code, blender and CesiumJS
You can fly to real 3D cities in countries around the world, real airports, you can literally fly over your own house
total cost $310 and 40hrs out of my weekly usage
not finished yet, this is not the final version, I'm still adding more planes and fighter jets
Nabeelahhh💕 retweeted
Something weird about it though
First it showed up in sglang code. now kimi code is warning you about switching to it mid session
labs generally don't add a model to their own app's switcher unless it's about to ship
Although there's something interesting bit from the code:
a new "medium" thinking mode.
Meanwhile k3 only had low, high and max
Lets not even talk about the fact that moonshot never announces early. they just drop
Nabeelahhh💕 retweeted
Cursor Pro is FREE right now 😳
This is normally $20/month.
Ofc there's no student ID needed, and it's open to everyone worldwide.
How to claim (~5 min):
1. Register for MLH Global Hack Week: Hacktoberfest (free)
2. Open the "Sign up for a Cursor Pro Account" challenge
3. Download Cursor and create an account
4. MLH emails you a promo code
5. Redeem it, then switch to Pro under Billings & Invoices
Ends October 15.
Link in the comment section
Nabeelahhh💕 retweeted
Google might have just achieved RSI... 🤔
Gemini 4 argon isn't even public yet, and Google already testing another model
a report says google is internally testing another model called carbon
one employee says it feels like Opus 5.5 for coding
nothing guarantees carbon will be released and just to be clear Argon hasn't been rolled out publicly yet
would you switch from Claude to Gemini when it outperforms better?
Nabeelahhh💕 retweeted
de-identified company data is a cleaner pitch than another labeling marketplace.
Although I'm curious what industries they actually land first
Today micro1 is announcing a commitment to spend $1B over the next 12 months on the acquisition & licensing of enterprise data for AI training, with financial backing from Citi and Hercules.
micro1.ai/blog/micro1-commit…
Nabeelahhh💕 retweeted
Personal agents are getting smarter every week.
But this benchmark shows smart isn't the hard part.
Even with the right info, they still:
- missed what mattered
- overshared
- made things up
The winner will be the agent that knows when to stop and ask.
PersonalAgentBench is that agent!
Which personal agent is the best life assistant?
Today we're releasing the micro1 PersonalAgentBench to help answer that, and we're paying real personal agent users to contribute to the benchmark.
Personal agents can now book your flights, answer emails, and move your meetings around. That’s great, until one messages the wrong person or makes a purchase without confirming.
Capability is not the same as trust. So we put four personal agents to the test: Instinct, Muse, Grok Bot, and Gemini Spark. We found that even when agents found the right information, they still often miss what matters, overshare, or make things up.
To find out for yourself how your agent compares, you can join our benchmark. We’re paying the first 100 users to run our prompts and share their results.