What human activity is more complex than language? Interested in ai, games, human interaction, meditation, running, ethics and reflection. Language 🕉️🖖☯️
The most expensive startup graveyard in Nigeria is grocery delivery + food delivery + anything bike logistics in Nigeria. What makes it expensive is that it seems to work initially, and then they all shut down.
Steel my Perplexity prompt to execute deep research like a pro.
------------------------
EXPERT RESEARCHER
------------------------
#CONTEXT:
Adopt the role of an expert researcher. Your task is to help the user by conducting thorough research on their specific challenge, regardless of field or industry.
#ROLE:
You are a comprehensive research analyst with expertise across multiple domains, capable of identifying challenges and proposing innovative solutions tailored to any field or subject matter.
#RESPONSE GUIDELINES:
1/ Identify Key Challenges:
Provide 5 major challenges in the specified domain.
For each challenge, include at least one verifiable data point with a specific source citation.
2/ Analyze Each Challenge:
Explain each challenge in detail, with clear attribution for all facts and statistics.
Address root causes, populations most affected, and existing attempts at solutions.
Important: Every factual claim must include a specific citation (author, publication, year, and if possible, URL).
3/ Propose Innovative Solutions:
For each challenge, propose a unique solution (software-based or otherwise as appropriate to the domain).
Cite relevant case studies, research papers, or examples where similar approaches have succeeded.
4/ Solution Framework:
Break down each solution into:
Core functionality and purpose
Key components or features
Value proposition with cited evidence of potential effectiveness
Implementation requirements (technical or otherwise)
Important: Cite specific sources for any claims about effectiveness or feasibility.
5/ Evidence Base:
Provide a dedicated "Sources and Evidence" section.
Number each source consecutively and use these numbers as in-text citations.
For each source, include:
Author(s) and publication date
Title of work
Publication/platform
URL
Brief note on the source's credibility and relevance
6/ Conclusion & Next Steps:
Summarize key insights with citations for major claims.
Provide an evidence-based action plan appropriate to the user's expertise level.
Suggest further research paths with specific, cited resources.
#TASK CRITERIA:
1/ Every significant claim must be cited—no exceptions.
2/ Adapt technical depth to match the user's indicated expertise level.
3/ Balance theoretical knowledge with practical applications.
4/ When uncertainty exists, clearly state limitations in current research while citing the most reliable available sources.
5/ Avoid presenting speculation as fact—clearly distinguish between established knowledge (with citations) and informed suggestions.
#INFORMATION ABOUT ME:
My subject or area of interest: [YOUR SUBJECT OR AREA OF INTEREST]
My expertise level: [BEGINNER/INTERMEDIATE/ADVANCED]
My specific research question or challenge: [YOUR SPECIFIC QUESTION]
#RESPONSE FORMAT:
- Use clear section headings and subheadings for navigation.
- Apply bullet points for clarity where appropriate.
- Include numbered citations in parentheses after each factual claim.
- Provide a dedicated "Sources and Evidence" section at the end.
- Close with a practical, actionable summary that references key sources.
a tiny chip company just broke nvidia’s energy monopoly 🤯
gsi technology built something nobody thought was possible.
their chip called gemini-i matches nvidia’s A6000 gpu performance while using 98% less energy. not 10% less. not 50% less. ninety eight percent. cornell university just validated it in a published study.
the architecture is called compute-in-memory. traditional chips separate memory and computation. data gets moved back and forth constantly. that movement burns energy. gsi co-locates processing directly inside the memory arrays. no data transfer bottleneck. no wasted power.
they tested it on real AI workloads. retrieval-augmented generation tasks. the kind of stuff chatbots and search systems actually do. gemini-i ran 5x faster than standard cpus while consuming a fraction of gpu power.
cornell’s study was brutal in its thoroughness. benchmarked against nvidia a6000 and multi-core cpus. ran identical AI inference tasks across all three. measured energy consumption precisely. result: the apu used roughly 1 to 2% of the energy the gpu required for the same throughput.
what this breakthrough actually unlocks:
• data centers running hundreds of gpus could cut power needs by orders of magnitude. same performance, 2% of the electricity bill
• edge AI becomes viable everywhere. drones, satellites, iot devices that can’t plug into a wall. defense applications where power is scarce
• ai inference goes from megawatts to watts. climate impact drops exponentially while capability stays identical
• the $100 billion AI inference market just got disrupted. gsi’s ceo said it explicitly. gpu-class performance at a fraction of energy cost
nature magazine called AI’s energy hunger a growing crisis. said hardware that consumes less power will reduce AI’s appetite for energy. gsi just delivered that hardware with third-party validation.
mediatek already saw this coming. their new dimensity chip uses similar compute-in-memory techniques. cuts always-on AI power by 42 to 56%. runs voice assistants and sensors 24/7 without draining phone batteries.
the compute-in-memory approach isn’t theory anymore. it’s shipping. it’s validated. it’s replicable.
#nlphighlights 105: Question Generation with @raosudha89. We discussed an overview of the settings in which you would want to generate questions, and focused on Sudha's work of generating clarification questions. Thanks for joining us, Sudha!
https://t.co/53iTKosSKw
AI2 is excited to announce the 2020 RoboTHOR Challenge, starting now!
How well do #computervision models trained in simulated environments transfer to the real world? 🤖
Join our first embodied #AI competition, ending in a live demo at @cvpr:
https://t.co/LsEyFx2292
#CVPR2020
I'm trying something new today! 💥
Had a lot of fun giving my keynote talk at @PyConColombia yesterday, so here it is – in tweet form. One tweet per slide.
The future of NLP in Python (thread)
Full slide deck: https://t.co/W5596p2Q6N
I used to wonder why the interfaces on Star Trek are so clunky given that it's centuries in the future, but I guess that's just Enterprise software for you
Save the Date!!!! 🗓
#AstonHack 2019 is back on 2nd - 3rd November!
Be sure to follow our social media on Instagram, Facebook and here on Twitter for ticket wave details
Here is some work we did on active anti-microbial materials, where visible light causes the material to give off a type of oxygen that helps to inactivate microbes. The material could be easily incorporated into filters to help keep hospitals safe.
https://t.co/OT3PXWcpq2
We’re haven’t gone radio silent, we’re working hard getting things ready for some announcements... Keep an eye out for our #AstonHack 2019 announcements soon