New a16z’s thesis: AI will unbundle the BPO and disrupt the $300b outsourcing market
Enterprises often outsource important, but high-volume and repetitive work to BPOs. These BPOs rely heavily on human labor, often leading to slow turnaround times, human error, and unsatisfactory results.
With modern AI, there's a clear opportunity to productize the BPO and unbundle the use cases they have dominated for decades.
BPOs are enormous businesses, with revenues in the tens of billions of dollars annually.
They serve customers across all major industries and do the necessary, but mundane, work their customers do not want to do themselves, such as customer support and claims processing.
This work can be inconsistent, seasonal, and involve high employee turnover, and managing it is not the core competency of most enterprises. For many, outsourcing this work was the logical solution.
a16Z believe key AI developments—including better language models, voice AI, and emerging browser capabilities—mean AI agents can now productize much of what BPOs do and enable enterprises to bring that work back in-house.
AI agents work at the speed of software, run 24/7, and can adapt to any culture and language with minimal human input. And importantly, they’re scalable, meaning they can actually expand markets by making it economically viable for companies to service more products, customers, and use cases than was possible before.
Enterprises often outsource important, but high-volume and repetitive work to BPOs. These BPOs rely heavily on human labor, often leading to slow turnaround times, human error, and unsatisfactory results.
With modern AI, there's a clear opportunity to productize the BPO and unbundle the use cases they have dominated for decades.
BPOs are enormous businesses, with revenues in the tens of billions of dollars annually.
They serve customers across all major industries and do the necessary, but mundane, work their customers do not want to do themselves, such as customer support and claims processing.
This work can be inconsistent, seasonal, and involve high employee turnover, and managing it is not the core competency of most enterprises. For many, outsourcing this work was the logical solution.
a16Z believe key AI developments—including better language models, voice AI, and emerging browser capabilities—mean AI agents can now productize much of what BPOs do and enable enterprises to bring that work back in-house.
AI agents work at the speed of software, run 24/7, and can adapt to any culture and language with minimal human input. And importantly, they’re scalable, meaning they can actually expand markets by making it economically viable for companies to service more products, customers, and use cases than was possible before.
#DeepSeek released recommended settings for the best experience:
• No system prompt
• Temperature: 0.6
• Official prompts for search & file upload
• Guidelines to mitigate model bypass thinking
• No system prompt
• Temperature: 0.6
• Official prompts for search & file upload
• Guidelines to mitigate model bypass thinking
GitHub
Add fileupload and websearch prompt by DeepSeekPH · Pull Request #399 · deepseek-ai/DeepSeek-R1
Here are prompts we use in DeepSeek Web and App
👍2
GREEN is a new lightweight neural network designed to how we analyze brain activity
GREEN combines two powerful approaches:
1. Wavelet-based frequency filtering to capture dynamic brain rhythms
2. Riemannian geometry to decode spatial patterns in EEG signals
EEG data holds secrets to brain health, cognitive function, and aging—but traditional methods often miss nuanced signals. GREEN changes the game by:
- Detecting subtle brain activity changes with exceptional sensitivity
- Operating on a low computational budget (ideal for real-world applications!)
- Delivering interpretable outputs to link findings to brain mechanisms.
GitHub.
GREEN combines two powerful approaches:
1. Wavelet-based frequency filtering to capture dynamic brain rhythms
2. Riemannian geometry to decode spatial patterns in EEG signals
EEG data holds secrets to brain health, cognitive function, and aging—but traditional methods often miss nuanced signals. GREEN changes the game by:
- Detecting subtle brain activity changes with exceptional sensitivity
- Operating on a low computational budget (ideal for real-world applications!)
- Delivering interpretable outputs to link findings to brain mechanisms.
GitHub.
Patterns
GREEN: A lightweight architecture using learnable wavelets and Riemannian geometry for biomarker exploration with EEG signals
Our brains generate electrical signals whose unique patterns reveal information about
aging, health, and even thoughts. This information is useful for studying brain disorders
and developing novel therapies. However, traditional analysis methods can miss…
aging, health, and even thoughts. This information is useful for studying brain disorders
and developing novel therapies. However, traditional analysis methods can miss…
👍3❤2
OpenAI released its Reasoning model best practices
1. Use delimiters for clarity: Use delimiters like markdown, XML tags, and section titles to clearly indicate distinct parts of the input, helping the model interpret different sections appropriately.
2. The guide differentiates between reasoning models (e.g., o1, o3-mini) and GPT models (e.g., GPT-4o).
Reasoning models are built for complex, multi-step tasks—such as planning, detailed document analysis, and visual interpretation—while GPT models focus on speed and cost efficiency for well-defined tasks.
- In practice, reasoning models excel at clarifying ambiguous prompts, extracting key details from extensive unstructured data, and performing multi-step planning or code review.
- They are best used with clear, concise prompts that include explicit constraints and delimiters;
- elaborate chain-of-thought instructions are unnecessary since these models reason internally.
1. Use delimiters for clarity: Use delimiters like markdown, XML tags, and section titles to clearly indicate distinct parts of the input, helping the model interpret different sections appropriately.
2. The guide differentiates between reasoning models (e.g., o1, o3-mini) and GPT models (e.g., GPT-4o).
Reasoning models are built for complex, multi-step tasks—such as planning, detailed document analysis, and visual interpretation—while GPT models focus on speed and cost efficiency for well-defined tasks.
- In practice, reasoning models excel at clarifying ambiguous prompts, extracting key details from extensive unstructured data, and performing multi-step planning or code review.
- They are best used with clear, concise prompts that include explicit constraints and delimiters;
- elaborate chain-of-thought instructions are unnecessary since these models reason internally.
Openai
Reasoning best practices | OpenAI API
Explore best practices for using o-series reasoning models, like o1 and o3-mini, vs. GPT models—including use cases, how to choose a model, and prompting guidance.
Google DeepMind released a short course on AGI safety
The course offers a concise and accessible introduction to AI alignment problems and our technical & governance approaches, consisting of short recorded talks and exercises (75 minutes total).
The course offers a concise and accessible introduction to AI alignment problems and our technical & governance approaches, consisting of short recorded talks and exercises (75 minutes total).
Medium
Introducing our short course on AGI safety
We are excited to release a short course on AGI safety for students, researchers and professionals interested in this topic. The course…
🆒1
#DeepSeek introduced CodeI/O, a new method that helps AI learn reasoning patterns hidden in code
Models train to predict inputs and outputs of given code, all while explaining its reasoning with Chain-of-Thought (CoT) in natural language.
CodeI/O improves models' general reasoning skills, such as:
- planning steps logically
- searching for solutions
- breaking problems into smaller parts
DeepSeek gathered over 810,000 Python code files from different sources to cover a big variety of reasoning styles, like puzzles, math problems, etc.
Then they cleaned and structured it into a unified format using DeepSeek-V2.5, ran the code and collected input-output pairs.
CODEI/O++: Improving training with multi-step feedback
It's an improved dataset where the model learns not just from its successes but also from its mistakes. To create it, researchers used a feedback loop:
- If the model gets an answer wrong, it's told why it was wrong and tries again.
- If the function fails to run, researchers include that feedback too.
- The model then revises its response and tries again.
This extra training step makes models even more accurate.
Data and models.
Models train to predict inputs and outputs of given code, all while explaining its reasoning with Chain-of-Thought (CoT) in natural language.
CodeI/O improves models' general reasoning skills, such as:
- planning steps logically
- searching for solutions
- breaking problems into smaller parts
DeepSeek gathered over 810,000 Python code files from different sources to cover a big variety of reasoning styles, like puzzles, math problems, etc.
Then they cleaned and structured it into a unified format using DeepSeek-V2.5, ran the code and collected input-output pairs.
CODEI/O++: Improving training with multi-step feedback
It's an improved dataset where the model learns not just from its successes but also from its mistakes. To create it, researchers used a feedback loop:
- If the model gets an answer wrong, it's told why it was wrong and tries again.
- If the function fails to run, researchers include that feedback too.
- The model then revises its response and tries again.
This extra training step makes models even more accurate.
Data and models.
arXiv.org
CodeI/O: Condensing Reasoning Patterns via Code Input-Output Prediction
Reasoning is a fundamental capability of Large Language Models. While prior research predominantly focuses on enhancing narrow skills like math or code generation, improving performance on many...
👍11
Krutrim LLM: Multilingual Foundational Model for over a Billion People
Krutrim LLM a 2 trillion token multilingual foundation model designed to serve Indian demographic needs through equitable representation of the country’s array of native tongues.
Training data incorporates the largest known Indic language dataset, mitigating associated data scarcity obstacles that encumber model parity across dialects.
Evaluations demonstrate Krutrim’s strong performance on Indic language benchmarks, surpassing or at par with SOTA models despite being significantly smaller in training flops.
Paper.
Krutrim LLM a 2 trillion token multilingual foundation model designed to serve Indian demographic needs through equitable representation of the country’s array of native tongues.
Training data incorporates the largest known Indic language dataset, mitigating associated data scarcity obstacles that encumber model parity across dialects.
Evaluations demonstrate Krutrim’s strong performance on Indic language benchmarks, surpassing or at par with SOTA models despite being significantly smaller in training flops.
Paper.
arXiv.org
Krutrim LLM: Multilingual Foundational Model for over a Billion People
India is a diverse society with unique challenges in developing AI systems, including linguistic diversity, oral traditions, data accessibility, and scalability. Existing foundation models are...
Google added infinite memory to Gemini, allowing it to remember & refer to past interactions while answering.
It's available in Gemini Advanced and can be tweaked by editing/deleting chats.
OpenAI is also working on a similar feature, but no release yet.
It's available in Gemini Advanced and can be tweaked by editing/deleting chats.
OpenAI is also working on a similar feature, but no release yet.
Google
Reference past chats for more tailored help with Gemini Advanced.
Starting today, Gemini can now recall your past chats to provide more helpful responses. Whether you’re asking a question about something you’ve already discussed, or as…
Mistral announced Saba a first regional language model
Mistral Saba is a 24B parameter model trained on meticulously curated datasets from across the Middle East and South Asia.
Mistral Saba supports Arabic and many Indian-origin languages, and is particularly strong in South Indian-origin languages such as Tamil and Malayalam.
Mistral Saba is a 24B parameter model trained on meticulously curated datasets from across the Middle East and South Asia.
Mistral Saba supports Arabic and many Indian-origin languages, and is particularly strong in South Indian-origin languages such as Tamil and Malayalam.
Step-Audio is 130 billion parameter multimodal LLM that is responsible for understanding and generating human speech.
huggingface.co
Step-Audio - a stepfun-ai Collection
Step-Audio model family, including Audio-Tokenizer, Audio-Chat and TTS
🔥4
The next DeepSeek is StepFun
They just dropped two huge SoTA and commercial-grade models on Huggingface
- T2V: 30B text to video model
- Audio chat, 130B audio understanding/generation model
- mit/Apache 2 license!
Model links on HF.
Demo for video generation.
They just dropped two huge SoTA and commercial-grade models on Huggingface
- T2V: 30B text to video model
- Audio chat, 130B audio understanding/generation model
- mit/Apache 2 license!
Model links on HF.
Demo for video generation.
Telegram
All about AI, Web 3.0, BCI
Step-Audio is 130 billion parameter multimodal LLM that is responsible for understanding and generating human speech.
🔥5🆒3
xAI announces Grok 3. Here is everything you need to know
Elon mentioned that Grok 3 is an order of magnitude more capable than Grok 2.
Total GPUs: 200K
The capacity was doubled in 92 days!
All of this compute was used to improve Grok -- which has lead to Grok 3.
Grok 3 involved 10x more training than Grok 2!
Grok finished pretraining in early January!
The model is still training.
Here are the benchmark numbers:
Grok 3 significantly outperforms other models in its category such as Gemini 2 Pro and GPT-4o. Even Grok-3 mini shows to be competitive.
Results of early Grok 3 in the Chatbot Arena (LMSYS)
It reached an Elo score of 1400 which no other model has achieved.
The model score keeps improving.
Grok 3 also has reasoning capabilities too!
The Grok team has been testing these capabilities which they have unlocked using RL.
The model is good, especially in coding.
Grok 3 coding example:
Thinking traces as generated as the model tries to solve the problem.
Elon confirmed that the thinking steps have been obscured to avoid getting copied.
Grok 3 also excels at creative coding like generating creative and novel games.
Elon emphasized Grok 3's creative emergent capabilities.
You can also use the Big Brain mode to use more compute and reasoning with Grok 3.
Grok 3 Reasoning performance:
The results correspond to the beta version of Grok-3 Reasoning.
It outperforms o1 and DeepSeek-R1 when given more test-time compute (allowing it to think longer).
The Grok 3 mini reasoning model is also very capable.
Grok 3 Reasoning Beta performance on AIME 2025.
Grok 3 shows generalization capabilities.
It not only does coding and math problem-solving, but it can also do other creative and useful real-world tasks.
One of the results generated with Grok 3 mini.
Bejeweled Tetris generated by Grok 3.
Grok 3 cannot only unlock test-time compute, it also enables capable agents.
These capabilities have led to a new product called DeepSearch.
"Next generation of search agents to understand the universe".
More on DeepSearch:
- the model can think deeply about user intent
- what facts to consider
- how many websites to browse
- it can cross-validate different sources.
DeepSearch also exposes the steps that it takes to conduct the search itself.
Improvements will happen rapidly and almost daily according to the team.
There is also a Grok-powered voice app coming too -- about a week away!
Open-source approach:
The last version will be open-sourced when the most recent version is fully out.
After Grok 3 stable version is out, it is highly likely Grok 2 will be open-sourced. (within a few months).
SuperGrok dedicated app is also available with a polished experience.
Try on the web as well: grok.com
The web will include the latest Grok features.
Elon mentioned that Grok 3 is an order of magnitude more capable than Grok 2.
Total GPUs: 200K
The capacity was doubled in 92 days!
All of this compute was used to improve Grok -- which has lead to Grok 3.
Grok 3 involved 10x more training than Grok 2!
Grok finished pretraining in early January!
The model is still training.
Here are the benchmark numbers:
Grok 3 significantly outperforms other models in its category such as Gemini 2 Pro and GPT-4o. Even Grok-3 mini shows to be competitive.
Results of early Grok 3 in the Chatbot Arena (LMSYS)
It reached an Elo score of 1400 which no other model has achieved.
The model score keeps improving.
Grok 3 also has reasoning capabilities too!
The Grok team has been testing these capabilities which they have unlocked using RL.
The model is good, especially in coding.
Grok 3 coding example:
Thinking traces as generated as the model tries to solve the problem.
Elon confirmed that the thinking steps have been obscured to avoid getting copied.
Grok 3 also excels at creative coding like generating creative and novel games.
Elon emphasized Grok 3's creative emergent capabilities.
You can also use the Big Brain mode to use more compute and reasoning with Grok 3.
Grok 3 Reasoning performance:
The results correspond to the beta version of Grok-3 Reasoning.
It outperforms o1 and DeepSeek-R1 when given more test-time compute (allowing it to think longer).
The Grok 3 mini reasoning model is also very capable.
Grok 3 Reasoning Beta performance on AIME 2025.
Grok 3 shows generalization capabilities.
It not only does coding and math problem-solving, but it can also do other creative and useful real-world tasks.
One of the results generated with Grok 3 mini.
Bejeweled Tetris generated by Grok 3.
Grok 3 cannot only unlock test-time compute, it also enables capable agents.
These capabilities have led to a new product called DeepSearch.
"Next generation of search agents to understand the universe".
More on DeepSearch:
- the model can think deeply about user intent
- what facts to consider
- how many websites to browse
- it can cross-validate different sources.
DeepSearch also exposes the steps that it takes to conduct the search itself.
Improvements will happen rapidly and almost daily according to the team.
There is also a Grok-powered voice app coming too -- about a week away!
Open-source approach:
The last version will be open-sourced when the most recent version is fully out.
After Grok 3 stable version is out, it is highly likely Grok 2 will be open-sourced. (within a few months).
SuperGrok dedicated app is also available with a polished experience.
Try on the web as well: grok.com
The web will include the latest Grok features.
👍8
Ilya Sutskever’s Startup Is Fundraising at $30 Billion-Plus Valuation
Ilya Sutskever is raising more than $1 billion for his startup at a valuation of over $30 billion, vaulting the nascent venture into the ranks of the world’s most valuable private technology companies.
Greenoaks Capital Partners is leading the deal for the startup, Safe Superintelligence, and plans to invest $500 million.
Greenoaks is also an investor in AI companies Scale AI and Databricks Inc.
The round marks a significant valuation jump from the $5 billion that Sutskever’s company was worth before. The financing talks are ongoing and the details could still change.
The company previously raised money from investors including Sequoia Capital and Andreessen Horowitz.
SSI focuses on developing safe AI systems. It isn’t generating revenue yet and doesn’t intend to sell AI products in the near future.
“This company is special in that its first product will be the safe superintelligence, and it will not do anything else up until then,” Sutskever told Bloomberg in June. “It will be fully insulated from the outside pressures of having to deal with a large and complicated product and having to be stuck in a competitive rat race.”
Sutskever was a key figure in the ouster of OpenAI Chief Executive Officer Sam Altman in 2023 before he helped Altman return.
Ilya Sutskever is raising more than $1 billion for his startup at a valuation of over $30 billion, vaulting the nascent venture into the ranks of the world’s most valuable private technology companies.
Greenoaks Capital Partners is leading the deal for the startup, Safe Superintelligence, and plans to invest $500 million.
Greenoaks is also an investor in AI companies Scale AI and Databricks Inc.
The round marks a significant valuation jump from the $5 billion that Sutskever’s company was worth before. The financing talks are ongoing and the details could still change.
The company previously raised money from investors including Sequoia Capital and Andreessen Horowitz.
SSI focuses on developing safe AI systems. It isn’t generating revenue yet and doesn’t intend to sell AI products in the near future.
“This company is special in that its first product will be the safe superintelligence, and it will not do anything else up until then,” Sutskever told Bloomberg in June. “It will be fully insulated from the outside pressures of having to deal with a large and complicated product and having to be stuck in a competitive rat race.”
Sutskever was a key figure in the ouster of OpenAI Chief Executive Officer Sam Altman in 2023 before he helped Altman return.
Bloomberg.com
OpenAI Co-Founder Sutskever’s Startup Is Fundraising at $30 Billion-Plus Valuation
OpenAI co-founder Ilya Sutskever is raising more than $1 billion for his startup at a valuation of over $30 billion, according to a person familiar with the matter — vaulting the nascent venture into the ranks of the world’s most valuable private technology…
❤3