Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

Introducing our latest technical report: Context Rot - How Increasing Input Tokens Impacts LLM Performance Our results reveal that models do not use their context uniformly. full report in replies

trychroma's profile picture

Chroma

30,214 subscribers

186,454 Aufrufe • vor 1 Jahr •via X (Twitter)

Wissenschaft & Technologie
LIVE

Anya Rossi• Live Now

Private livecam show

11 Kommentare

Profilbild von Chroma
Chromavor 1 Jahr

Read the full report here:

Profilbild von Mobile Scanner
Mobile Scannervor 1 Jahr

Scan any documents, convert images into text, PDF files, etc. 👍

Profilbild von jason liu
jason liuvor 1 Jahr

wow veo3 is so good

Profilbild von Kinjal Nandy
Kinjal Nandyvor 1 Jahr

Lfg @kellyhongsn

Profilbild von noah
noahvor 1 Jahr

feel like we've all known this so im glad it was rigerously tested

Profilbild von dinos
dinosvor 1 Jahr

you guys are on fire

Profilbild von Chroma
Chromavor 1 Jahr

spread the news

Profilbild von Allan Ryan
Allan Ryanvor 1 Jahr

Is she AI?

Profilbild von Chroma
Chromavor 1 Jahr

100% human intelligence

Profilbild von sarv
sarvvor 1 Jahr

Wooo @kellyhongsn!!

Profilbild von Olivier
Oliviervor 1 Jahr

benchmark different context engineering strategies next good content

Ähnliche Videos

Today we released our Q4 and full year 2025 financial results, showing performance in line with our expectations. Our President and CEO, Justin Hotard takes us through some of the key takeaways. Read full report: #QComms #TeamNokia
1:30

Today we released our Q4 and full year 2025 financial results, showing performance in line with our expectations. Our President and CEO, Justin Hotard takes us through some of the key takeaways. Read full report: #QComms #TeamNokia

Nokia

27,924 Aufrufe • vor 6 Monaten

An important report by Sky News’s Becky Johnson that highlights the increasing number of children educated in unregulated educational settings and the perverse incentives that act against our most inclusive schools and nurseries. Full report below. NAHT
1:01

An important report by Sky News’s Becky Johnson that highlights the increasing number of children educated in unregulated educational settings and the perverse incentives that act against our most inclusive schools and nurseries. Full report below. NAHT

Simon Kidwell

28,426 Aufrufe • vor 2 Jahren

New short course: LLMs as Operating Systems: Agent Memory, created with Letta, and taught by its founders Charles Packer and Sarah Wooders. An LLM's input context window has limited space. Using a longer input context also costs more and results in slower processing. So, managing what's stored in this context window is important. In the innovative paper MemGPT: Towards LLMs as Operating Systems, its authors (which include the instructors) proposed using an LLM agent to manage this context window. Their system uses a large persistent memory that stores everything that could be included in the input context, and an agent decides what is actually included. Take the example of building a chatbot that needs to remember what's been said earlier in a conversation (perhaps over many days of interaction with a user). As the conversation's length grows, the memory management agent will move information from the input context to a persistent searchable database; summarize information to keep relevant facts in the input context; and restore relevant conversation elements from further back in time. This allows a chatbot to keep what's currently most relevant in its input context memory to generate the next response. When I read the original MemGPT paper, I thought it was an innovative technique for handling memory for LLMs. The open-source Letta framework, which we'll use in this course, makes MemGPT easy to implement. It adds memory to your LLM agents and gives them transparent long-term memory. In detail, you’ll learn: - How to build an agent that can edit its own limited input context memory, using tools and multi-step reasoning - What is a memory hierarchy (an idea from computer operating systems, which use a cache to speed up memory access), and how these ideas apply to managing the LLM input context (where the input context window is a "cache" storing the most relevant information; and an agent decides what to move in and out of this to/from a larger persistent storage system) - How to implement multi-agent collaboration by letting different agents share blocks of memory This course will give you a sophisticated understanding of memory management for LLMs, which is important for chatbots having long conversations, and for complex agentic workflows. Please sign up here!
2:57

New short course: LLMs as Operating Systems: Agent Memory, created with Letta, and taught by its founders Charles Packer and Sarah Wooders. An LLM's input context window has limited space. Using a longer input context also costs more and results in slower processing. So, managing what's stored in this context window is important. In the innovative paper MemGPT: Towards LLMs as Operating Systems, its authors (which include the instructors) proposed using an LLM agent to manage this context window. Their system uses a large persistent memory that stores everything that could be included in the input context, and an agent decides what is actually included. Take the example of building a chatbot that needs to remember what's been said earlier in a conversation (perhaps over many days of interaction with a user). As the conversation's length grows, the memory management agent will move information from the input context to a persistent searchable database; summarize information to keep relevant facts in the input context; and restore relevant conversation elements from further back in time. This allows a chatbot to keep what's currently most relevant in its input context memory to generate the next response. When I read the original MemGPT paper, I thought it was an innovative technique for handling memory for LLMs. The open-source Letta framework, which we'll use in this course, makes MemGPT easy to implement. It adds memory to your LLM agents and gives them transparent long-term memory. In detail, you’ll learn: - How to build an agent that can edit its own limited input context memory, using tools and multi-step reasoning - What is a memory hierarchy (an idea from computer operating systems, which use a cache to speed up memory access), and how these ideas apply to managing the LLM input context (where the input context window is a "cache" storing the most relevant information; and an agent decides what to move in and out of this to/from a larger persistent storage system) - How to implement multi-agent collaboration by letting different agents share blocks of memory This course will give you a sophisticated understanding of memory management for LLMs, which is important for chatbots having long conversations, and for complex agentic workflows. Please sign up here!

Andrew Ng

200,950 Aufrufe • vor 1 Jahr

Introducing DuoAttention: Our new framework slashes both memory and latency for long-context LLMs without sacrificing performance! By applying full KV cache only to critical heads, we achieve: ⚡ 2.55x memory reduction ⚡ 2.18x decoding speedup ⚡ 3.3M tokens on a single A100 GPU
1:00

Introducing DuoAttention: Our new framework slashes both memory and latency for long-context LLMs without sacrificing performance! By applying full KV cache only to critical heads, we achieve: ⚡ 2.55x memory reduction ⚡ 2.18x decoding speedup ⚡ 3.3M tokens on a single A100 GPU

Guangxuan Xiao

31,085 Aufrufe • vor 1 Jahr

Most conversational AI understands words, not people. Introducing Raven-1, our audio and video perception model that gives AI the ability to understand emotion, intent, and context the way humans do.
4:50

Most conversational AI understands words, not people. Introducing Raven-1, our audio and video perception model that gives AI the ability to understand emotion, intent, and context the way humans do.

Tavus

1,672,610 Aufrufe • vor 6 Monaten

Our 2026 ESA Essential Facts report is LIVE! 🎮 212.3M Americans play video games—older players are rising, most parents who play do so with their kids, and much more. Read the full report & latest industry insights: #ESAEssentialFacts
0:26

Our 2026 ESA Essential Facts report is LIVE! 🎮 212.3M Americans play video games—older players are rising, most parents who play do so with their kids, and much more. Read the full report & latest industry insights: #ESAEssentialFacts

Entertainment Software Association

68,046 Aufrufe • vor 2 Monaten

📹 Is it worth submitting a video of a close pass? Footage can help, but context matters. Our latest blog explains how Operation SNAP works and why a short description can be as important as the video itself. 👉 Make your report count:
0:36

📹 Is it worth submitting a video of a close pass? Footage can help, but context matters. Our latest blog explains how Operation SNAP works and why a short description can be as important as the video itself. 👉 Make your report count:

Cycling UK

11,535 Aufrufe • vor 7 Monaten

MiniMax M2.7 at home running on 4x DGX Sparks vLLM serving full BF16 weights, 200k context OpenCode having the model monitor its own hardware and report thermals, tokens/sec, TTFT, and other runtime stats in real time What benchmarks / workflows / things do you wanna see next?
1:24

MiniMax M2.7 at home running on 4x DGX Sparks vLLM serving full BF16 weights, 200k context OpenCode having the model monitor its own hardware and report thermals, tokens/sec, TTFT, and other runtime stats in real time What benchmarks / workflows / things do you wanna see next?

Ahmad

92,491 Aufrufe • vor 4 Monaten

Our latest speech-to-speech model is faster, more accurate, and excels at function calling. Watch @promptshant and Brian Fioca build a realtime voice agent that can search the web and hand off tasks to reasoning models with full context.
1:58

Our latest speech-to-speech model is faster, more accurate, and excels at function calling. Watch @promptshant and Brian Fioca build a realtime voice agent that can search the web and hand off tasks to reasoning models with full context.

OpenAI Developers

81,822 Aufrufe • vor 1 Jahr

Our last Stanford guest lecture - Ekdeep Singh Lubana on what counts as an explanation & a neuro-inspired "model systems approach" to interp Plus, how in-context learning and many-shot jailbreaking are explained by LLM representations changing in-context (as a case study for that approach) 00:33 - What counts as an explanation? 04:47 - Levels of analysis & standard interpretability approaches 18:19 - The "model systems" approach to interp [Case study on in-context learning] 23:36 - How LLM representations change in-context 44:10 - Modeling ICL with rational analysis 1:10:54 - Conclusion & questions Thanks again to Surya Ganguli for having us in his class!
1:18:26

Our last Stanford guest lecture - Ekdeep Singh Lubana on what counts as an explanation & a neuro-inspired "model systems approach" to interp Plus, how in-context learning and many-shot jailbreaking are explained by LLM representations changing in-context (as a case study for that approach) 00:33 - What counts as an explanation? 04:47 - Levels of analysis & standard interpretability approaches 18:19 - The "model systems" approach to interp [Case study on in-context learning] 23:36 - How LLM representations change in-context 44:10 - Modeling ICL with rational analysis 1:10:54 - Conclusion & questions Thanks again to Surya Ganguli for having us in his class!

Goodfire

31,410 Aufrufe • vor 8 Monaten

Girls as young as 5 don't feel that they belong in sport. Our groundbreaking research uncovers the devastating truth of how young girls are having their self-belief undermined. The full report can be found on our website: #girlsbelonginsport #IWD2023
1:45

Girls as young as 5 don't feel that they belong in sport. Our groundbreaking research uncovers the devastating truth of how young girls are having their self-belief undermined. The full report can be found on our website: #girlsbelonginsport #IWD2023

Women in Sport

83,000 Aufrufe • vor 3 Jahren

Our Q4 and full-year 2022 results show that we grew 11% in Q4 year-on-year, and 6% over the full-year in constant currency. Our full year comparable operating margin was stable 12.5%. Watch Pekka Lundmark explain the highlights and read the report here:
2:07

Our Q4 and full-year 2022 results show that we grew 11% in Q4 year-on-year, and 6% over the full-year in constant currency. Our full year comparable operating margin was stable 12.5%. Watch Pekka Lundmark explain the highlights and read the report here:

Nokia

50,746 Aufrufe • vor 3 Jahren

Watch as we interview our team about their old Black Friday purchases & savings, calculate what they'd be in Bitcoin, and reveal the surprising results! 🤯 Use our calculator to find out how much your own savings could have grown! #TrezorBlackFriday
1:18

Watch as we interview our team about their old Black Friday purchases & savings, calculate what they'd be in Bitcoin, and reveal the surprising results! 🤯 Use our calculator to find out how much your own savings could have grown! #TrezorBlackFriday

Trezor

16,281 Aufrufe • vor 1 Jahr

Our co-founder Pratyush Kumar highlighted Sarvam's "contrarian bet" of being a full-stack AI platform. That means we build every layer: the models, how they run (inference), how agents work with each other (orchestration), and the AI applications people use. This lets us build AI that actually works for India's unique context and diverse language needs.
1:30

Our co-founder Pratyush Kumar highlighted Sarvam's "contrarian bet" of being a full-stack AI platform. That means we build every layer: the models, how they run (inference), how agents work with each other (orchestration), and the AI applications people use. This lets us build AI that actually works for India's unique context and diverse language needs.

Sarvam AI

16,310 Aufrufe • vor 8 Monaten

❓Fave nugget from our chat w GitHub CEO Thomas Dohmke on our latest Sequoia Capital Training Data episode with me & Sonya Huang 🐥: On what timeframe do you think AI will be as good as an average developer, and a 10x developer? 1⃣ Models today are able to write better code than the average developer, if prompted in the right way and given the right context 2⃣ But models today lack creativity, critical thinking, and systems thinking... Models today still need to learn how to decompose problems and make the right technical decisions
3:28

❓Fave nugget from our chat w GitHub CEO Thomas Dohmke on our latest Sequoia Capital Training Data episode with me & Sonya Huang 🐥: On what timeframe do you think AI will be as good as an average developer, and a 10x developer? 1⃣ Models today are able to write better code than the average developer, if prompted in the right way and given the right context 2⃣ But models today lack creativity, critical thinking, and systems thinking... Models today still need to learn how to decompose problems and make the right technical decisions

Stephanie Zhan

22,187 Aufrufe • vor 2 Jahren

ICE whistleblower Ryan Schwank warned Congress that, without reforms, "ICE will graduate thousands of new officers who do not know their constitutional duty, do not know the limits of their authority and who do not have the training to recognize an unlawful order." Our report:
1:45

ICE whistleblower Ryan Schwank warned Congress that, without reforms, "ICE will graduate thousands of new officers who do not know their constitutional duty, do not know the limits of their authority and who do not have the training to recognize an unlawful order." Our report:

Camilo Montoya-Galvez

94,571 Aufrufe • vor 6 Monaten

Following up on our Intelligence Index v4.1 release yesterday, in the video below, Daniel from our team shares a short overview of what's changed: 1. Three upgraded evaluations: Terminal-Bench 2.1, τ³-Bench Banking and GDPval-AA v2 2. Cost, time, and tokens per task: Understand the cost, time, and tokens of tasks across our Index and for individual evals, and how these trade off against Intelligence 3. Cached input token reporting: We now report the amount of cached tokens a particular model uses and how this influences cost
2:40

Following up on our Intelligence Index v4.1 release yesterday, in the video below, Daniel from our team shares a short overview of what's changed: 1. Three upgraded evaluations: Terminal-Bench 2.1, τ³-Bench Banking and GDPval-AA v2 2. Cost, time, and tokens per task: Understand the cost, time, and tokens of tasks across our Index and for individual evals, and how these trade off against Intelligence 3. Cached input token reporting: We now report the amount of cached tokens a particular model uses and how this influences cost

Artificial Analysis

12,491 Aufrufe • vor 2 Monaten

With the Commissioner's final report in our hands, the question echoes: What's our next move? How do we navigate this pivotal moment? We turn the question to you: What will you do about it? Beyond a mere document, the report stands as a powerful tool and a testament to our history, poised to shape our future. Its potential for transformation lies not solely within its pages but within the hands of the Canadian people. In our citizens' toolkit, we now wield a robust instrument to demand change. The question that lingers: How will you unleash its power? Download the report: Learn how you can Take Action: #NCI #CommissionersReport #DemandChange #CanadianCitizens #Canada #cdnpoli #COVID19 #TruthMatters
2:24

With the Commissioner's final report in our hands, the question echoes: What's our next move? How do we navigate this pivotal moment? We turn the question to you: What will you do about it? Beyond a mere document, the report stands as a powerful tool and a testament to our history, poised to shape our future. Its potential for transformation lies not solely within its pages but within the hands of the Canadian people. In our citizens' toolkit, we now wield a robust instrument to demand change. The question that lingers: How will you unleash its power? Download the report: Learn how you can Take Action: #NCI #CommissionersReport #DemandChange #CanadianCitizens #Canada #cdnpoli #COVID19 #TruthMatters

National Citizens Inquiry (NCI | CeNC)

16,848 Aufrufe • vor 2 Jahren