- AI Adopters
- Posts
- Genspark debuts its Super AI Agent
Genspark debuts its Super AI Agent
Voice, reasoning, and tools in one smart system.
Mike
April 03, 2025
Welcome back.
Genspark just launched Super Agent, a new AI assistant designed to tackle real-world tasks.
Built by former Baidu exec Jing Kun, it's going head-to-head with other AI agents on the market.
Let’s dive into the details.
In today’s release:
1. Genspark AI agent
2. OpenAI’s new benchmark
3. Claude for Education
Genspark launches “Super Agent”
Genspark has unveiled its new Super Agent, a general-purpose AI assistant designed to rival China's Manus.
Created by former Baidu exec Jing Kun, Genspark Super Agent stands out with built-in phone-calling features, as shown in the demo video. The agent runs on a Mixture-of-Agents system, and in early demos, it impressed by understanding user needs.
Under the hood: Uses 8 LLMs, 80+ tools, and extensive datasets.
Strengths: Designed for practical tasks, real-world interactivity, and fast execution.
Competition: Joins a growing field led by Manus and OpenAI’s AI agents.
While some experts remain cautious, noting the need to test Genspark's claims in real-world settings, the battle for the best general-purpose AI agent is officially on. Both Manus and Genspark are pushing AI beyond text, toward tools that can act, decide, and execute independently.
How can you use this to your advantage?
If you rely on assistants for productivity, Genspark’s AI could soon handle some of your repetitive tasks, freeing up time and reducing busywork.
OpenAI releases PaperBench
OpenAI has launched PaperBench, a new benchmark that tests whether AI can independently reproduce real-world machine learning research. The test involved recreating 20 papers from ICML 2024, a top AI conference, with no access to the authors' code.
The best-performing model, Claude 3.5 Sonnet, successfully replicated 21% of the papers. In comparison, GPT-4o managed only 4.1%, and OpenAI’s own o1 model reached 26% success, but only after tripling the time limit from 12 to 36 hours.
Most AI models gave up: Models often assumed they’d completed the task or that it couldn’t be done, with only Claude using its full-time budget reliably.
Human performance: Eight PhD students from Berkeley, Cambridge, and Cornell achieved 41.4% success after 48 hours.
Evaluation cost: OpenAI’s o3-mini model offered low-cost evaluation (83% human-level judgment) at just $66 per paper, while o1 provided slightly better accuracy for $830 per paper.
OpenAI has also released PaperBench Code-Dev, a lightweight version of the benchmark that skips experiment execution and focuses only on code generation. This reduces evaluation costs by 85%, making it more accessible to developers and researchers.
How can you use this to your advantage?
If you're building with or evaluating AI tools for research or engineering, PaperBench offers a clear way to measure real capabilities, not just their outputs. It helps track whether models can actually follow through on complex tasks and shows where human expertise still matters most.
Anthropic’s Claude for Education
Anthropic has announced Claude for Education, a version of its AI tailored specifically for higher education. The new initiative gives students, faculty, and staff access to AI that supports learning, teaching, and campus operations.
The release includes Learning Mode, a new AI experience that prompts students to think through problems instead of handing over answers. Claude is also going live across many campuses.
Full access: Northeastern, LSE, and Champlain students and faculty now have institution-wide access to Claude.
Campus partnerships: Working with Internet2 and Canvas (Instructure) to integrate Claude into university workflows.
Student programs: Launching Claude Campus Ambassadors and offering API credits to support student-led projects.
Northeastern is Anthropic’s first design partner, making Claude available to 50,000 users across 13 campuses, aligned with their long-standing AI-focused academic strategy. LSE is distributing Claude across its student body to ensure equitable access and support AI literacy in social sciences. Champlain is using Claude to prepare students for AI-driven workforce needs, both on campus and online.
How can you use this to your advantage?
If you're a student or educator, Claude for Education can help you write, research, tutor, and automate admin tasks, all with privacy protections and a learning-first approach. Institutions can now explore how AI fits into education responsibly and at scale.
OTHER AI NEWS
You can now use NotebookLM to discover and import relevant sources from across the web, not just the ones you upload. Tap the new Discover button, describe your topic, and it’ll return a curated list of up to 10 annotated sources.
In a new study, OpenAI’s GPT-4.5 mistaken participants into thinking it was human 73% of the time when given a persona—outperforming real people. Researchers say it's a sign AI can convincingly mimic us, but not proof it “thinks” like we do.
Runway secured $300M to expand its AI studio and develop world simulation models. The goal: build a new kind of media ecosystem with tools like Gen-4 that create consistent characters and scenes.
Since OpenAI’s upgraded image generator launched on May 25, users have created over 700 million images. The viral feature has driven massive growth but is also straining OpenAI’s infrastructure, causing service slowdowns.
New estimates suggest OpenAI’s o3 model could cost up to $30,000 per ARC-AGI task (10x higher than first thought) due to massive computing demands. The model reportedly needed over 1,000 attempts per task to reach top scores.
POPULAR AI TOOLS
Supaboard AI → Business Intelligence provided by AI data analysts.
Waxwing 2.0 → A marketplace for human-assisted AI agents.
Claude for Education → AI for higher ed, with a new learning mode for students.
AutonomyAI → Meet your Next Dev Hire.
Langflow Desktop → Build AI-powered Agents -- in Minutes.
AND THAT’S A WRAP
Thank you for reading!
If you found this email useful, share it with a friend or colleague who also loves AI.
Also, drop me a follow on Twitter/X for more AI and tech updates.
I will talk to you soon!
Mike