|
Hey,
Your daily briefing is ready. You can finally take a break from the AI firehose.
Our algos spent the night splitting signal from noise and pulled the top news, models, papers, and repos.
Here’s the must-read:
|
|
Together with:
|
|
|
|
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer.
Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
|
|
Top News
|
|
|
|
Perplexity launches Computer, an AI system that runs full workflows end-to-end instead of answering single prompts
|
|
20,125 Likes
|
|
Perplexity just launched Perplexity Computer, a general-purpose AI system that runs full workflows instead of answering single prompts. It researches, designs, codes, deploys, and manages projects end-to-end.
The key shift: it orchestrates multiple frontier models in parallel, not just one.
Perplexity Computer acts like a digital co-worker. You describe an outcome. It breaks that into tasks, spawns sub-agents, and runs them in isolated compute environments with real browser access, file systems, and tool integrations.
Key features and results
-
Multi-model orchestration across 19 frontier models
-
Parallel sub-agents for research, coding, and deployment
-
Persistent memory across sessions and projects
-
Real browser, filesystem, and tool integrations
-
Usage-based pricing with model selection and spending caps
-
10,000 monthly credits for Max users + 20,000 launch bonus
How to use it
-
Open Perplexity (Max subscription required).
-
Describe a goal, not just a prompt.
-
Optionally choose models or set token limits.
-
Let agents execute asynchronously.
-
Track progress or run multiple projects at once.
|
|
|
|
Presented by WorkOS
|
|
The Fastest Growing AI Companies Choose WorkOS
|
|
AI companies are scaling faster than ever and hitting enterprise requirements far earlier in their lifecycle.
SSO, SCIM, RBAC, audit logs, compliance. These are non-negotiable for enterprise customers, and expensive and difficult to build and maintain in-house.
That’s why OpenAI, Anthropic, Cursor, Vercel, and more use WorkOS to power authentication, identity, and security. It's the infrastructure layer that makes applications enterprise ready without slowing teams down.
|
|
|
|
partner with us
|
|
Top News
|
|
|
|
Alibaba ships open source Qwen 3.5 medium, proving a 35B model can outperform a 235B predecessor with smarter architecture
|
|
6,621 Likes
|
|
Alibaba just released the Qwen 3.5 Medium series, and the message is clear: scaling parameters alone no longer defines progress.
Instead of pushing model size, this release focuses on architecture design, higher-quality training data, and RL. RL means the model improves through feedback, similar to reward-based training.
The highlight is Qwen3.5-35B-A3B, a Mixture-of-Experts (MoE) model. MoE activates only a small subset of parameters per token, which reduces compute cost while preserving performance.
Although it has 35B total parameters, it uses ~3B active parameters per step and surpasses the older 235B predecessor on several benchmarks.
Key Features
- 1M token context (Flash default)
- Built-in tool use for agent workflows
- FP8 open weights
- Native vLLM and SGLang support
Results
- 35B-A3B surpasses prior 235B model
- Strong performance on coding and agent benchmarks
- Reduced inference cost via sparse activation
|
|
|
|
Presented by Lambda
|
|
Build custom AI models in hours, not months
|
|
Lambda and Oumi have partnered to deliver end-to-end custom model development and deployment in hours, with better accuracy, lower latency, and stronger privacy built in.
-
Automated test set creation, evaluation, and fine-tuning
-
Deployment on secure, single-tenant infrastructure
-
70% cost savings and 20% quality gains in one healthcare deployment
|
|
|
|
partner with us
|
|
Top Paper
|
|
|
|
Ai2 releases PreScience, an open benchmark that tests whether AI can forecast how scientific research unfolds end-to-end
|
|
1,027 Likes
|
|
Can AI predict how science unfolds, not just one paper but the whole process? Ai2 introduces PreScience, built with University of Chicago and supported by National Science Foundation, to test exactly that. It evaluates whether models can forecast research from team formation to real-world impact.
PreScience uses ~100K real AI papers from arXiv and splits prediction into four chained tasks:
-
Collaborator prediction: Forecast who works together next
-
Prior work selection: Predict which papers they cite
-
Contribution generation: Generate the title and abstract
-
Impact prediction: Forecast first-year citations
Ai2 also introduces LACERScore, an LLM-as-judge metric that scores abstract alignment on a 1-10 scale. GPT-5 scores 5.6/10. Simple co-author heuristics beat complex ML for collaboration prediction. Highly cited papers remain hardest to forecast.
When Ai2 simulates 12 months of AI research, generated papers cluster into narrower idea spaces than real research. The bottleneck appears in the generation step.
You access the dataset, code, and evaluation scripts via GitHub and Hugging Face.
|
|
|
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We’re passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.
Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We’re based in the United States. By using our site, you agree to be governed by U.S. laws.
|
|
|
Looking to promote your company, product, service, or event to 250,000+ AI developers?
|
|
|
|
|