Alex L (@a_levitator) header
Alex L (@a_levitator) profile picture

AI thought follower @ramplabs

New York, NY
Joined January 2025
Photo by Alex L (@a_levitator)Photo by Alex L (@a_levitator)Photo by Alex L (@a_levitator)Photo by Alex L (@a_levitator)Photo by Alex L (@a_levitator)Photo by Alex L (@a_levitator)Photo by Alex L (@a_levitator)Photo by Alex L (@a_levitator)Photo by Alex L (@a_levitator)Photo by Alex L (@a_levitator)
Ramp Labs can't keep avoiding the obvious elephant in the room
We’re publishing Ramp SWE-Bench v2, featuring 500 tasks built from work shipped to production by over 200 of our engineers, more rigorous verification, and frontier models running in their native harnesses. labs.ramp.com/swebench
2
30
24,112
Me on Opposite Day:
Brute force computation is not enough for intelligence. Clever new methods are needed. The science of AI is far from finished. IMHO
1
242
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
2,367
7,258
5,755
55,924
10,080,878
Ramp inspect is the best internal tool I have ever used, and now we are sharing it with the world! Please use it, you will not regret 📈
1
16
551
Cyber whitepill: 1. Economics favor defenders. Coding costs money now, and the quality/security of your code is a function of how much money you are willing to spend on it. Attackers must also invest large amounts of money on inference, and attack success rate scales with compute spend. Large software corporations & the US govt have much more money to burn than the vast majority of attackers. 2. Critical infrastructure has always been vulnerable to physical and digital threats. Powerful nation-state adversaries avoid large cyberattacks against critical US infra not because they are unable to penetrate, but because of fear of retaliation. AI does not remove this deterrent. 3. For closed-source software, defenders have privileged source code access. White-box vuln hunting is much easier & more compute-efficient than black-box. 4. Labs are giving their most powerful models to defenders. Many of the most complex & destructive issues are only discoverable by frontier intelligence which is exclusively accessible to defenders. 5. The blue team has AI too. Capable defenders will use agents to patch vulns, harden systems, and respond to incidents faster. There's a lot of worry and sensationalism about cyber right now, some of it justified. Important to remember that defenders have an asymmetric advantage, and can maintain it with consistent & ambitious action.
5
1
1
19
1,415
Despite what haters on this website are saying, my mid-50s parents are using instinct to book their flights and absolutely love it
1
10
548
This is a skill issue from devs. Have to stay ambitious & keep pushing every new model to its limit
OpenAI, Anthropic and other leading AI labs face a growing challenge: Better models aren't necessarily attracting more usage. A 10-point intelligence advantage now earns just 20% more developer traffic, down from nearly 69% in late 2025. Full story: thein.fo/4hGJ9jY
1
7
592
Pomegranate seeds bring joy into my worthless incel life
2
13
634
If your CTO isn’t stimming in the office ur ngmi
1
11
565
Biggest Ramp news of the day
We had early access to @NVIDIAAI Nemotron 3.5 Super VL and post-trained it for one of our highest-volume workflows: receipt extraction. One epoch of supervised fine-tuning on 10,000 examples made it competitive with leading open and closed-source models on our receipt benchmark.
4
34
2,529
Who is @tryramp? in math: my solution ➗ in history: my king 👑 in art: my muse 🎨 in science: my oxygen 💨 in geography: my world 🌎
Today, Ramp raised $1.85B at a $60B valuation. Seven years ago, Karim and I wrote our mission on a Post-it: “Save companies time and money.” Our mission hasn't changed. It’s just gotten bigger.
1
22
758
It's actually crazy how sampling more tokens with a bigger model is the solution to 99% of applied ai problems
1
3
424
It’s time to lock in
2
238
I’m actually really glad AI exists because otherwise stuff like this would be the big news in our industry
Your binder is about to get a major upgrade. Rip packs on your phone. Every pull backed by a real, physical card. Vault it or ship it. Soon on Coinbase.
1
14
3,800