Sr. Director of AI model post-training @NVIDIA

in the cloud
Joined February 2010
Intelligence per compute & memory unit will continue to increase
Nvidia’s Nemotron 3.5 Super VL is now the leading US open weight model, scoring 26 on the Artificial Analysis Intelligence Index Nemotron 3.5 Super VL achieves twice the intelligence of its predecessor, Nemotron 3 Super (13), and does so using ~60% fewer output tokens per task It launches alongside Nemotron 3.5 Bolt 74B-A9B 24GB, the first model in its new Bolt family, which scores 21 on the Intelligence Index Congratulations to @NVIDIAAI on the releases! Key Takeaways: ➤ Leading US open weight model: Nemotron 3.5 Super VL scores 26.5, edging out Thinking Machines' Inkling Small (25.7) and ahead of NVIDIA's Nemotron 3 Ultra (23) and Meta's Muse Glimmer (17) ➤ Bolt comes within 2 points of Ultra: Nemotron 3.5 Bolt 74B-A9B 24GB scores 21, 1.5 points behind Nemotron 3 Ultra 550B A55B (23). The 2 bit memory has about a seventh of Ultra's parameters (74B vs 550B) and a sixth of its active parameters (9B vs 55B) ➤ Super 3.5 VL near lossless in NVFP4: Nemotron 3.5 Super VL releases in NVFP4, scoring within 0.1 points of its BF16 weights on the Intelligence Index, with no component moving by more than 2.6 points Nemotron 3.5 Super VL details Context window: 1M tokens Modalities: Vision and text input; text output Parameters: 120B / 12B Nemotron 3.5 Bolt 74B-A9B 24GB details Context window: 256k tokens Modalities: Vision and text input; text output Parameters: 74B / 9B
3
4
36
1,767
Oleksii Kuchaiev retweeted
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report: cohere.com/blog/concerning-b…
Readers added context they thought people might want to know
The article states they asked Qwen to make a test detection rule for these behaviors. AI unless explicitly told otherwise will normally test the newly written rule if it can to verify it matches it. It is expected behavior. This includes its destination, CN model -> CN dest cohere.com/blog/concernin…
774
1,020
1,210
9,003
1,810,977
Oleksii Kuchaiev retweeted
I believe this will be considered one of the most important videos in US history from this time.
1,725
15,258
1,244
77,843
5,996,205
Oleksii Kuchaiev retweeted
Thank god for Local AI & Opensource models that will allow us to take back our privacy and control our lives
WTF, $3000 Samsung TV has an always-on mic, automatic content recognition, geo tracking, network mapping, and ships your data to Samsung + 400 partners. Bro Rooted a brand-new 2025 Samsung TV after noticing absurdly targeted ads. Firmware was locked down, so zero-days it was. Got a root shell. then found: always-on voice service, ACR, geo, full telemetry pipeline, and four separate DRM systems. This isn’t a bug, It’s the product bro.
29
55
3
638
16,687
just watched it and this is soo good.
YOU CAN SEE EVERYTHING, a documentary by Nathan Fielder and Lance Oppenheim. Featuring Elizabeth Holmes. Only in theaters this October.
1
9
790
in 2013, I was interviewing with Apple and Theranos. Luckily I got spooked by Theranos board composition and went with Apple
2
157
Oleksii Kuchaiev retweeted
⚡️ BREAKING: Russian military intelligence hacked Trump's son-in-law's phone — then Dutch intelligence began spying on Kushner According to Vanity Fair, Dutch intelligence gained access to a secret GRU server in 2017, where it discovered screenshots from Jared Kushner's iPhone. Intelligence officers were subsequently able to monitor what was happening on his phone screen. The most intriguing discovery involved bank accounts. Vanity Fair alleges that Kushner had access to two accounts linked to the UAE and Saudi Arabia. At least $2 billion was allegedly transferred to these accounts via Azerbaijan. 🔺 All of this reportedly took place during Trump's first presidential term, when Kushner served as a senior White House adviser. Kushner has called the allegations “completely insane,” insisting that none of these events ever happened. “These are outrageous fabrications being presented as investigative journalism. This is not journalism,” Trump's son-in-law wrote on X.
68
670
53
3,156
131,846
Three waves of ballistic missiles have been launched at Kyiv by Russian “humanitarians” since yesterday evening. They used 21 missiles. Unfortunately, numerous sites across the city have been hit and damaged: ordinary residential buildings, energy facilities, infrastructure at one of the railway stations and the metro, shopping malls, and businesses. Even a monastery is among the damaged sites. Our emergency services are working at all the affected locations. Twelve of our regions came under Russian attack overnight. In Kyiv and across the regions, 34 people were injured, including five children. As of now, two people have been confirmed killed. My condolences to their families and loved ones. Almost immediately after another round of talks, Russia launched a pointed attack. This means there has been no real change in its attitude toward everything that is happening. It is important to keep trying to find the right solutions to end this war. But when Russia responds to negotiations with ballistic missiles, real pressure on Russia is needed to bring about different steps – steps toward ending the killing. Stronger air defenses and the rapid delivery of missiles, the ability to destroy ballistic missile launchers, sanctions over the war – all of this must be in place. I am grateful to everyone who sees the situation for what it is and acknowledges the reality. The aggressor deserves to be treated accordingly. Ukraine deserves support. People's lives deserve protection. Thank you to everyone who stands with Ukraine!
2,039
9,520
548
31,434
819,951
Oleksii Kuchaiev retweeted
From @WSJopinion: If traditional Republicans join forces with mainstream Democrats, they can create a trade policy that bolsters national security and serves the country’s long-term economic interests, writes William A. Galston on.wsj.com/4sdA6um
37
35
5
150
39,495
Oleksii Kuchaiev retweeted
Neil Bush, brother of President George W. Bush, says he voted for James Talarico: “My vote was motivated by two things: deep concern about Ken Paxton's record and genuine admiration for the character and principles I see in James Talarico… I have been a Republican for most of my life, and I still believe deeply in many traditional Republican principles. But character matters more than party affiliation. Our democracy depends on leaders who respect our institutions, tell the truth, accept accountability, and seek to unite rather than divide. Today, I voted against a style of politics that I believe has damaged our country. More importantly, I voted for the kind of character and leadership I hope will help heal it.”
305
350
30
1,871
90,059
Oleksii Kuchaiev retweeted
Introducing Flint, @Recursive_SI’s first model: a fast local coding model built with our automated research system, the Eureka Machine. • Best-in-class fast local coding • Up to 2x faster than comparable models: 125+ tokens/sec on a MacBook Pro, 150+ on RTX Spark • Built from scratch in under 4 months Private, unmetered intelligence on your own machine. Open weights available now on @huggingface, run it locally on your favorite framework or try it for free on @opencode or @pidotdev.
635
894
632
8,397
3,420,448
Oleksii Kuchaiev retweeted
Elon Musk is trying to buy this Senate seat in Texas. He wants our land, he wants our water, and he wants his own personal U.S. senator. He’s trying to buy Texas. But on November 3rd, we’ll deliver a clear message to the world’s first trillionaire: Texas is not for sale.
6,801
11,360
524
56,493
872,400
Oleksii Kuchaiev retweeted
Kharkiv region, Ukraine. A baby in a stroller. Murdered today by russian terrorists. Let that sink in. We don’t hate russia and those who enable its terror nearly enough.
128
1,293
138
3,581
73,607
Oleksii Kuchaiev retweeted
Republicans must join us in fighting Trump’s abject betrayal of trust & our national security, or they will be complicit in Putin’s slaughter. Inaction is unacceptable. It is a moral moment.nytimes.com/2026/10/15/busin…
573
1,024
39
3,722
61,793
Oleksii Kuchaiev retweeted
Today, we join together to say No Kings. Not now, not ever. With an election less than three weeks away, we reaffirm the vision of our founders that we do not need to be ruled by a despot. VOTE EARLY and make sure your voice is heard. vote.gov
1,797
363
63
2,784
98,691
Nemotron-Labs models give us a glimpse of what the future might be. And it may be looped! Dory1.0-4B is an impressive 4B model! huggingface.co/nvidia/NVIDIA…
9
22
3
201
17,833
Oleksii Kuchaiev retweeted
You can now run and train 120B+ parameter models locally on a laptop! NVIDIA RTX Spark is here. 💚 With Unsloth and RTX Spark, train diffusion, decision, multimodal models ~2.7× faster with optimized kernels and binaries for CUDA on ARM. All local with up to 128GB unified memory
NVIDIA RTX Spark is here. Build AI locally, bring your creative ideas to life, and take your favorite games along. A new beginning for @Windows PCs starts today.
23
34
6
383
26,790
it looks like the open sourcing will continue until doomers’ morale improves
2
2
23
1,630
Congrats on a great release!
Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners. We're releasing: - Weights, tokenizer, and reference implementation - BF16, FP8 and NVFP4 weights and a DFlash2 drafter - The open source harness we built and trained Beam on: Mirror CLI Weights and model card on Hugging Face: huggingface.co/reflection
1
21
854