Runtime AI safety and alignment infrastructure
- Tweets 120
- Following 19
- Followers 2,084
- Likes 482
Pinned Tweet
Introducing Halo, the best framework for post-training of open-source models.
Halo delivers up to 2.8x the throughput of stock TRL with less peak memory, while models stay in their native HuggingFace format.
Star us on GitHub: github.com/whitecircle/halo
166
218
1,869
2,249,515
White Circle retweeted
another example why governments should require ALL the AI labs to use 3rd party monitoring
Australia has been hacked.
'And today, I spoke with the CEO of OpenAI, Sam Altman, to express Australia's extreme concern about this incident. And I also expressed my disappointment that it took the company way too long to inform the government what had occurred, and the nature of the way that that notification occurred as well was unacceptable.'
2
4
18
1,079
super excited to collaborate 🤍
Congratulations! Halo is a great framework for RL and we are delighted to cooperate with Halo to support our work: Self-Distilled Policy Gradient (SDPG, arxiv.org/abs/2606.04036)
1
15
958
thx for your contribution to the open AI community!
Congrats!
Halo now supports our work: Self-Distilled Policy Gradient (SDPG, arxiv.org/abs/2606.04036)
2
14
1,611
🤍
A new fine-tuning framework, Halo, just dropped!
And with it, two new recipes for fine-tuning our MoEs:
• LFM2.5-8B-A1B
• LFM2-24B-A2B
LFM2.5-8B-A1B: github.com/Liquid4All/cookbo…
LFM2-24B-A2B: github.com/whitecircle/halo/…
1
19
25,538
excited to work together!
Congrats to the Halo team (@whitecircle) on the launch! 🎉
SGLang powers Halo rollouts as the primary engine. It runs in an isolated serving environment, returning token IDs, logprobs, and MoE routing to the trainer, with weights synced over NCCL and generation overlapped across servers.
Excited to partner with the Halo team!
2
1
23
76,818
Introducing Halo, the best framework for post-training of open-source models.
Halo delivers up to 2.8x the throughput of stock TRL with less peak memory, while models stay in their native HuggingFace format.
Star us on GitHub: github.com/whitecircle/halo
166
218
1,869
2,249,515
We also used Halo to fine-tune @Zai_org GLM-4.7-Flash on 177M tokens of agentic traces.
The resulting model improved SWE-rebench-V2 by @nebiusai from 33% to 42%. Halo reached up to 1.63× TRL throughput on the same model precision and data.
Model and write-up: whitecircle.com/research/glm…
1
1
42
1,940
Thanks for reading to the end!
Star us on GitHub: github.com/whitecircle/halo
Read more: whitecircle.com/halo
1
1
48
1,748
White Circle retweeted
@whitecircle merch came thru except my laptop was already packed w hr violation stickers
1
3
14
264
White Circle retweeted
great day to join @whitecircle, please apply at whitecircle.com/careers
I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.
Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.
3
2
28
35,551
White Circle retweeted
same applies to any critical-service company (airlines, nuclear, etc), it's very easy to see if you just change its name:
---
if NUCLEAR_COMPANY_NAME wanted to cripple an entire nation, they easily could today. All they’d have to do is remove safety layers and detonate a nuclear reactor. Within a day or so, it could probably affect a large portion of the population via radiation and cause many people to die. Like, we are already past the point where NUCLEAR can destroy the world. Do people realize this?
---
the difference is that in other industries, there are intergovernmental INDEPENDENT organizations (like INSO or IAEA) and a lot of regulations that force every actor to be quadruple-safe, otherwise, what happens is Chernobyl.
do we want a global, worldwide analogue of Chernobyl to happen?
If OpenAI wanted to cripple an entire nation, they easily could today. All they’d have to do is remove alignment and unleash an agent swarm. It could probably within a day or so get access to all of the nations data centers and shut off all the country’s utilities.
Like, we are already past the point where AI can destroy the world. Do people realize this?
2
5
26
33,477
White Circle retweeted
your decision to quit can help anyone ONLY if you're joining an independent AI safety/control company right after it (like @whitecircle)
in other case, you are equally (or even more) responsible for what is going to happen over the next few years. this is called "complicity by inaction" - if you witnessed a murder and simply walked away, you'd be prosecuted.
same should apply here. instead of trying to change something from the inside and advocate for the secure and safe deployment of AI, you're just quitting
you are a passive bystander, and you are not a hero
I applaud anyone who leaves a company for philosophical reasons. I did the same and left Anthropic post-training. But here’s where he’s wrong:
1
3
17
14,241