Jamie Little (@AnalyticsJamie) header
Jamie Little (@AnalyticsJamie) profile picture

Global Headhunter - Managing Partner - Bateman Fox. Partnering with VCs/PEs & PortCos to land S-Tier talent. Snowboarding 🏂 | Diving 🤿 | Family Man.

SF | NYC | LDN
Joined May 2014
Photo by Jamie Little (@AnalyticsJamie)Photo by Jamie Little (@AnalyticsJamie)Photo by Jamie Little (@AnalyticsJamie)Photo by Jamie Little (@AnalyticsJamie)Photo by Jamie Little (@AnalyticsJamie)Photo by Jamie Little (@AnalyticsJamie)Photo by Jamie Little (@AnalyticsJamie)Photo by Jamie Little (@AnalyticsJamie)Photo by Jamie Little (@AnalyticsJamie)
Jamie Little retweeted
The first @Neuralink product is called Telepathy
8,001
8,029
2,917
82,010
24,982,385
Jamie Little retweeted
Thank you @ylecun for open source, LLaMa, and being a paragon of anti doomerism! 👏 #NeurIPS2023
9
8
1
263
36,026
Replying to @geomblog
1. Save interesting-sounding ones to a file called "To Read". 2. Never get around To Reading. 3. Three months later: Papers are out of date. Repeat forever.
7
9
131
5,957
Jamie Little retweeted
If you are wanting to read something over the weekend or want a quick information on how can agents be optimized, the below post is very useful 👇🏻
🚨New Paper Alert -🦙⚒️LLMCompiler If you enjoyed @karpathy's video on an "LLM OS", you're going to want to check out this paper (complete with a full open source repo!) This system compiles an effective plan for executing multiple tasks in parallel, letting LLM agents work together efficiently on complex tasks at a large scale Paper: arxiv.org/abs/2312.04511 Code: github.com/SqueezeAILab/LLMC… Big s/o to @sehoonkim418 @snrpsnr @ryan_tabrizi @nicholaszlee Michael Mahoney @KurtKeutzer @amir__gholami
3
10
3,170
Jamie Little retweeted
The LLMCompiler paper takes the concept of an ‘LLM OS’ to new heights, showcasing how LLM agents can effectively collaborate on complex, large-scale tasks. A game-changer for anyone interested in AI’s potential in multi-tasking and efficiency. The open-source aspect is a huge plus! #AI #MachineLearning #OpenSource – at Texas, USA
🚨New Paper Alert -🦙⚒️LLMCompiler If you enjoyed @karpathy's video on an "LLM OS", you're going to want to check out this paper (complete with a full open source repo!) This system compiles an effective plan for executing multiple tasks in parallel, letting LLM agents work together efficiently on complex tasks at a large scale Paper: arxiv.org/abs/2312.04511 Code: github.com/SqueezeAILab/LLMC… Big s/o to @sehoonkim418 @snrpsnr @ryan_tabrizi @nicholaszlee Michael Mahoney @KurtKeutzer @amir__gholami
9
20
3,944
Jamie Little retweeted
How do you increase a 27% accuracy score to 98%? By adding “Here is the most relevant sentence in the context:” to your prompt. Yesterday, Anthropic published a blog post outlining how they achieved significantly better results on the same evaluation by adding this sentence to the start of Claude’s response. Prompt engineering is a fascinating topic, and I am curious about what we will know a year from now. Read the full article here: anthropic.com/index/claude-2…
39
181
39
1,453
447,812
Gemini, Google's new generative Al models to rival ChatGPT won't be available in Europe for now. Regulatory hurdles are to blame. The @GoogleDeepMind team have been working hard and would love to get it shipped in Europe. What are your thoughts when it comes to Al regulation in Europe? Is it blocking innovation and causing the continent to lag behind other parts of the world? Read more here: lnkd.in/gUSaavtF #ai #artificialintelligence #deeptech #startups #euregulations
1
224
LLM? old news. New hotness will be LVMs (Large Vision Models). This time, looks like @GoogleAI is finally staying ahead of @OpenAI. How well does it work? a 🧵-> 1/n arxiv.org/pdf/2302.05442.pdf
16
83
4
657
204,958
Jamie Little retweeted
⭐️ Multi-modal RAG public benchmark ⭐️ Multi-modal LLMs are one of the most important emerging trends for the coming year. They promise to unlock new types of Q+A assistants over visual content, but multi-modal RAG architectures remain uncertain. We are releasing a small public benchmark, which will be extended over time, to help with evaluation of multi-modal RAG apps. The benchmark is a set of Q+A pairs on a slide deck (@datadoghq investor presentation) that require reasoning about the visual slide content. We used the benchmark to test 2 approaches for multi-modal RAG: (1) Embed slides as images w/ multi-modal embeddings, retrieve slides relevant to a question, pass images to GPT-4V for answer (2) Use GPT-4V to summarize slides, embed the text summaries, use summaries to retrieve images, pass images to GPT-4V for answer We found: (1) Multi-modal RAG approaches outperform text-only ingestion by a wide margin (2) Multi-modal embeddings may struggle to retrieve semantically similar slides, but simplify the architecture and likely have a high ceiling as multi-modal embeddings improve (3) Image summarization is very good for image retrieval, but suffers from higher cost of pre-computing image summaries and requires a complex compare retriever to stores summaries + images You can test these approaches for yourself following the documentation below, and we will continue to expand this benchmark over time. Documentation: langchain-ai.github.io/langc… Blog: blog.langchain.dev/multi-mod…
3
79
7
320
39,570
Jamie Little retweeted
If this is how promotions work at your company you are screwed
Replying to @scarletinked
I know this for a fact: If you ask a random HR Partner if a person is ready for promotion to Director, do you know what they're going to ask first? "How many people are reporting to them?" This system is broken.
7
2
1
61
27,618
I was speaking to a Talent Acquisition Leader recently who was struggling to hire. Positions had been open for 6+ months. One of the problems was not being able to compete on salary. They were slightly below the market rate. 1/13
1
1
384
Much progress being made on #AIAct here in Brussels! 22hrs of negotiation, starting at 3pm yesterday under the belt! Getting back into it at 9am tomorrow morning! Watch this space! #Trilogue
3
11
35
3,867
The Federal Reserve will cut interest rates 6 times in 2024 as the economy shows clear signs of cooling down, ING has said.
436
733
254
5,804
1,512,896
Jamie Little retweeted
The core tech of the current AI breakthroughs is as old as AI itself: the perceptron was already invented in 1957! How can we improve on this? Liquid.ai is a new MIT startup that rethinks function approximation, using Liquid Neural Networks: linkedin.com/posts/joschabac…
18
41
4
351
50,378
Google DeepMind's timeline to date for those who don't know - Did you know they partnered with Open AI twice? Here's we go: 2010 Co-founded by Demis Hassabis, Shane Legg, Mustafa Suleyman 2011 > Raised an undisclosed Series A from Founders Fund & Horizons Ventures. I think Elon Musk also invested at this time. 2012 > Facebook tries to acquire DeepMind but fails 2013 > DeepMind reveals having developed an AI algorithm able to learn how to play iconic early video games like Breakout and Pong simply by watching them being played on a vintage 1977 Atari 2600 games console. 2014 > Google engages in its largest European acquisition to date with a deal to buy DeepMind for around $500 million. From then on, DeepMind started becoming known worldwide. 2015 > DeepMind's AlphaGo beats the European Go champion Fan Hui, a 2 dan (out of 9 dan possible) professional, five to zero. This is the first time an artificial intelligence defeats a professional Go player. 2016 > DeepMind generates £40.3 million in revenue after their deep learning tool has been deployed in many environments across Google 2017 > DeepMind’s safety team partners with OpenAI in the development of an algorithm which can infer what humans want by being told which of two proposed behaviors is better. > DeepMind announces AlphaGo Zero, a software capable of mastering the Chinese game of Go without help from human players. > DeepMind recorded £54.4 million up 35% from £40.3 million in 2016. 2018 > DeepMind partners with OpenAI in a new paper that proposes a new method to train reinforcement learning agents in ways that enables them to surpass human performance. > DeepMind, along with tech leaders, including Elon Musk, sign a pledge promising to not develop “lethal autonomous weapons.” > DeepMind unveils AlphaFold, an algorithm able to predict the complex, three-dimensional shapes into which proteins can be folded. 2019 > DeepMind's AI AlphaStar is revealed to outperform human professionals at StarCraft II, beating the humans 10 games in a row. > DeepMind has also struggled to break even with its growing expenses, which include the computing necessary to train massive AI models and a large headcount (around 1,000 employees). Alphabet was forced to write off nearly £1.1 billion ($1.5 billion) of DeepMind’s debt. 2020 > DeepMind made a leap in the field of AI with its system, AlphaFold, which was recognized as a solution to the “protein-folding problem.” 2021 > Revenue rose to £889.4 million and profit stood at £102.4 million > Trained Gopher - A 280 billion parameter language model that was better than OpenAI's GPT-3 2022 > Created an AI system named AlphaCode that it says “writes computer programs at a competitive level.” 2023 > Google Brain merges with DeepMind as Alphabet plans to take on OpenAI's ChatGPT > Introduced PaLM 2, our next generation large language model that has improved multilingual, reasoning and coding capabilities. > Hello Gemini
4
18
2,268
"LLM Visualization" (bbycroft.net/llm?utm_source=…) JK--This is fascinating, mystifying, intimidating, or all of the above.
1
101
Jamie Little retweeted
Very excited by our new brain-robotic interface work @corl_conf — NOIR is a general-purpose, intelligent BRI system that enables humans to command robots, using non-invasive EEG signals. 🤩👇
Introducing our new work @corl_conf 2023, a novel brain-robot interface system: NOIR (Neural Signal Operated Intelligent Robots). Website: noir-corl.github.io/ Paper: arxiv.org/abs/2311.01454 🧠🤖
10
94
12
582
179,813
Jamie Little retweeted
How often should plans change as a Founder? 💚
plans should be measured in decades, execution should be measured in weeks. you gotta have both.
3
1
6
3,174