Ben Geist
706 posts
Ben Geist
@b_geist
Research Eng @ramplabs / physics + math nerd / Kate Bush fan
Brooklyn, NY Katılım Temmuz 2019
371 Takip Edilen626 Takipçiler
Pretty sure 50% of internal token spend is completely useless, but right now it's hard to know which 50%.
As an admin I'd love a dashboard that breaks down each person's spend into summarized clusters. Much easier to spend more when you can draw a clear line to value.
Ed Zitron@edzitron
Uber’s COO has said that it’s getting “harder to justify” its AI costs because there was no way to show a link between AI spend and any meaningful increase in useful features. This is the first time I’ve seen a company say this directly. businessinsider.com/uber-coo-andre…
English
A scaffold for modern memory systems could very well just be optimized block sparse attn
0x三井瘦@0xmitsui
看起来 @MiniMax_AI M3很快就要来了。工程负责人@SkylerMiao7 之前发的一个技术图中可以看到 MiniMax M3 模型确定将会有百万上下文，采用基于GQA(Grouped Query Attention)的动态块稀疏注意力设计。先用 Index Branch 做粗检索，再用 Sparse Branch 对选中的 block 做真实 attention，它的逻辑是：当前 query 不需要看全部历史，只需要看 top-k 相关历史块。打个比方就是看书时候不是把整本书每一页都重读，而是先快速查目录/索引，定位几个相关章节，再精读。这个设计的效果也很明显，一百万上下文，prefill比之前快9.7倍，decode快15.6倍。期待到时候看看DeepSeek V4 和 Minimax M3 谁才是性价比之王。
English
It was cool to work with the @GoogleDeepMind team on their Gemini Managed Agents service! They did a great job at integrating the agent runtime into the platform to quickly build production agents. Congrats on the launch!
English
I created a list of my favorite cafes to work at in NYC.
Coffee shops > offices.
I started my company inside a coffee shop.
Beautiful spaces. Good energy. Surrounded by people locking in. And you randomly meet the most interesting people.
My list includes the cafe name, neighborhood, and a proprietary, confidential scoring system based on:
- Work space (tables, outlets, WiFi, design)
- Food (quality, options, price)
- Music (playlist, can you take calls)
- People (do interesting people go here)
- Coffee (does it hit)
- Vibes (overall energy)
I'd love to share this list with you + add new spots.
Comment "CAFE" and I'll DM you the list.
English
English
We partnered with @PrimeIntellect to build Fast Ask, a small RL-trained subagent that helps our Sheets agent find answers in spreadsheets. It scores +4% over Opus on exact match accuracy at Haiku latency.
English
@henrytdowling @RampLabs @PrimeIntellect Yes the majority of effort was building the tasks and reward so that the training was stable prime rl made the rest of the process simplified
English
@RampLabs @PrimeIntellect this is really cool. was most of the work building the RL environment?
English
Ben Geist retweetledi
We worked with @RampLabs to train Fast Ask using Lab
A small RL-trained subagent that helps the Ramp Sheets agent find answers in spreadsheets.
The resulting FastAsk model outperformed Opus 4.6, while obtaining Haiku-level speeds at even lower costs.
Ramp Labs@RampLabs
English
time is money. save both.
was an absolute pleasure to collaborate with @ramplabs on this using `prime lab`
Ramp Labs@RampLabs
English
Worked with @PrimeIntellect’s new prime-rl stack to RL a subagent for Ramp Sheets. Latency and accuracy don’t have to be at odds with RL.
Ramp Labs@RampLabs
English
@a_levitator has been building self healing agentic systems. I’ve first hand seen their success, highly suggest going to see him speak about it!
Matt Turck@mattturck
AI folks in NYC -- Data Driven NYC (#121) this Tuesday at 6pm. Come meet fellow AI builders and our speakers: * @RampLabs has been cooking lately with a lot of agentic innovation; Alex Levinson will demo * @EstuaryDev provides unified data infra for AI - CEO @dyaffe RSVP: luma.com/ddnyc121
English