Super excited to have started at @GoodfireAI as an MTS today! Looking forward to researching what LLMs are up to under the hood 🔍 🐼
281132610K
Posts with #GoodFire
#GoodFire
Public posts on X that use #GoodFire.
Related hashtags
Happened upon a grass fire in Middle Bidwell Park today. We were there 5 minutes before the FD showed up. It was a textbook #goodfire, cleaning up thatch, and burning star thistle seeds, with easily controllable rates of spread. Prescribed fire doesn't have to be complicated!6774
AI.Engineer 🔜 Shanghai & SF CODE@aiDotEngineerNovember 10 at AIE CODE: Agents and Software Factories. A full day of talks alongside the workshops: • @KaranVaidya6 of @composio on taking agents from coding to knowledge work. • @frasergeorgew of @fivetran on giving agents the right context. • Utkarsh Shrivastava of @netflix on building a shared AI assistant beyond MCP. • @DanJBalsam of @GoodfireAI on detecting reward hacking inside the model. San Francisco. November 10 requires the workshop add-on. See the program and apply:
32132.9K- AI.Engineer 🔜 Shanghai & SF CODE@aiDotEngineerWe're excited to announce our AI Engineer Code Summit speaker lineup! → VP Gemini Post-Training Lead, @GoogleDeepMind CEO, @e2b CTO, @ExaAILabs CEO, @warpdotdev CEO, @aviator_co_ CEO, @fivetran CTO, @Lovable CTO, @iamovercut CTO, @earendilworks (Pi) CEO, @diffusionio CPO, @cognition CTO, @GoodfireAI MTS, @AnthropicAI CTO, @composio GPM, @Cloudflare MTS, @databricks CEO, @AmpCode Co-Founder, @gimletlabs CEO, @mixedbreadai CEO, @humanlayer_dev Head of AI, @MonacoGTM Lecturer, @UCBerkeley Principal Engineer, @SpaceXSI CEO, @TypeSafeAI (Jev) Founder, Bun at Anthropic VP Product, @reflection_ai Research Lead, Reflection AI Staff Software Engineer, @SonarSource Principal Software Engineer, @tryramp Senior Engineering Manager, @netflix Forward Deployed Engineer, @baseten Also joining us: @kevinhou22 — Google DeepMind Antigravity @Steve_Yegge — Creator of Beads & Gas Town @lucasmeijer — @kentcdodds — Software Engineer & Educator @trq212 — Independent @simonw — Django, Pelicans, Lethal Trifecta Presented by Google DeepMind. November 10–12 in San Francisco. Attendance is by application. →474419.2K
Forest Fire Facts@forestfirefactsWhy are #wildfires getting more intense? Check this story from #CentralOregon that shows it's actually due to a lack of #GoodFire. Read more at the link below.21018- Forest Fire Facts@forestfirefacts"We need a [Smokey] who can gain public support for using fire as a tool, a persuasive bear who can help cities near forests and rural towns prepare for living with wildfires." #ForestFireFacts #GoodFire #SmokeyBear7531446
- Forest Fire Facts@forestfirefactsOn the paradox of too much and not enough fire in the West: "It’s time for radical action—and radical responsibility." Read more: #GoodFire #ForestFireFacts
5634 - Burned about 500 acres today to reduce invasive annual grasses, boost native perennials, and improve prairie chicken habitat. With @pheasants4ever @Nature_Fire and others. #goodfire412103
Nick Alonso@Nick__AlonsoI joined @GoodfireAI! I'm excited to be part of the team and work toward their mission of using interpretability techniques to improve alignment and training. Check out this recent work for an example of their progress in this direction 👇Goodfire @GoodfireAIWe cut through the mystery around interpretability in training. A single, mathematically rigorous framework transforms model internals into training feedback and works across domains, problems, and modalities. We call this *mechanistic shaping*, and we’ve used it to mitigate reward hacking, remove shortcuts in chest x-rays, and reduce sycophancy and harmfulness. We believe this is a big step forward for technical alignment. But it carries a known risk: training against a model’s internals can make it harder to monitor. We explore both the technique and this concern below. 🧵22461.3K
Goodfire@GoodfireAIWe cut through the mystery around interpretability in training. A single, mathematically rigorous framework transforms model internals into training feedback and works across domains, problems, and modalities. We call this *mechanistic shaping*, and we’ve used it to mitigate reward hacking, remove shortcuts in chest x-rays, and reduce sycophancy and harmfulness. We believe this is a big step forward for technical alignment. But it carries a known risk: training against a model’s internals can make it harder to monitor. We explore both the technique and this concern below. 🧵311581.4K97.2K- Got to help put some #goodfire on the ground at the Big Chico Creek Ecological Reserve yesterday. Kudos to @Pyrogeographer for having the vision and championing a fire-forward land management program @chicostate!5660
- I just got an early Christmas present from the California State Fire Marshal! It's been a long time coming, and I'd like to shout out to all of the people who have mentored me and given me time with #GoodFire. I'm consulting under the umbrella of @DeerCrResources. Call us!
1341524.6K - Got to help put 10 acres of #goodfire on the ground today in the Butte County Foothills with a crew of locals. This is our 4th burn on this property, where we are building a toehold to do larger, more complex burns in an area with critical WUI wildfire hazards. #buttetrex.
27142 
Shaping and Monitoring Models Through Their Internal Representations latest mechanistic alignment research from @GoodfireAIGoodfire @GoodfireAIWe cut through the mystery around interpretability in training. A single, mathematically rigorous framework transforms model internals into training feedback and works across domains, problems, and modalities. We call this *mechanistic shaping*, and we’ve used it to mitigate reward hacking, remove shortcuts in chest x-rays, and reduce sycophancy and harmfulness. We believe this is a big step forward for technical alignment. But it carries a known risk: training against a model’s internals can make it harder to monitor. We explore both the technique and this concern below. 🧵115407- This is a good example of taking advantage of low-hanging fruit when it comes to prescribed fire. There are lots of places we can use fire to reduce heavy fuel loadings with minimal staffing and zero risk of escapes. #goodfireYosemite Fire and Aviation Management @YosemiteFire340 acres complete in Crane Flat area with work continuing 2/18. Our friends from Stanislaus came over too #jackpot #burning #likeagoodneighbor #rightconditions31083
Yosemite Fire and Aviation Management@YosemiteFire340 acres complete in Crane Flat area with work continuing 2/18. Our friends from Stanislaus came over too #jackpot #burning #likeagoodneighbor #rightconditions2967- My day as seen by a webcam operator. Laying down some #goodfire in the Butte County Foothills with 20 volunteers and some backup from @CALFIRE_ButteCo. We burned 7 acres of some of the lowest-elevation pine and black oak forest in the area, protecting this stand from future fires3530
- Still reviewing footage from our prescribed burn in #BidwellPark this week. Thinking this clip is one of the best I've ever captured. It shows lighters pulling 'flanking strips' into the wind to generate a slow-moving flame front to kill star thistle. #goodfire #ecologicalburning47877.1K
@RobertAvramMD@RobertAvramMD @GoodfireAI Got it, I'll send this to your Grok Bot. It'll land there in a clean, ready-to-use shape.00096- Forest Fire Facts@forestfirefactsPrescribed burns and ecological thinning helped protect the South Lake Tahoe area during the Caldor fire -- see how with these animated maps at the link below. #GoodFire #FireWise3835
- Goodfire@GoodfireAI@GoodfireAIThe full paper & blog post have more detail on our results and conclusions. Taken together, we think this opens up a set of tools and analyses contributing to safely solving technical alignment.361344.9K
- If we want to walk the talk about expanding the pace and scale of prescribed fire, we could light off fires like the #SitesFire under today's conditions and burn five to ten thousand acres at a time with pretty minimal risk to anything. Not all #GoodFire is low intensity!
412927.9K - Show more posts
