Akshat Gupta

AI First Centaur | Ex-@IgniteTech | Ex-@GoogleIndia | Ex-@gonoise

Disclaimer: Views are personal.

[Akshat Gupta @Akshat_G]

@TransportDelhi
Kindly issue me the IDP at the earliest which is stuck since 13-Jan-2026.

Application Number: 204785226

Name: Akshat Gupta

RTO: Sarai Kale Khan (DL3)

Status: Scrutiny - (Verification of Proof Documents)

Counter: RMSCR6

[🔥🗞The Informant @theinformant_x]

❗️🌋🇪🇹 - Ethiopia's Hayli Gubbi Volcano Awakens After 10,000 Years

In a stunning geological event, Ethiopia's Hayli Gubbi volcano—long dormant in the remote Danakil Depression of the Afar Rift—erupted explosively for the first time in recorded history on November 23, 2025.

The outburst, which began around 8:30 a.m. UTC, propelled a massive ash plume soaring up to 15 kilometers (about 50,000 feet) into the atmosphere, accompanied by significant sulfur dioxide emissions.

Located roughly 15 kilometers southeast of the perpetually active Erta Ale volcano, Hayli Gubbi's awakening marks its initial eruption in the Holocene epoch, with no prior activity documented in the last 10,000 years.

Satellite imagery from Planet Labs captured the onset in near real-time, revealing a dramatic plume that quickly dispersed eastward over the Red Sea, affecting air quality in parts of Yemen, Oman, and even drifting toward Iran, Pakistan, and India.

As of the latest update from the Toulouse Volcanic Ash Advisory Center (VAAC) on November 23 evening, the eruption has ceased, with observed volcanic ash limited to 10,000 feet (3,000 meters).

No damage to communities or disruptions to air travel have been reported, thanks to the volcano's isolated location in one of Earth's most extreme environments.

Scientists continue to monitor the site for potential renewed activity, highlighting the dynamic tectonics of the Afar Rift, a key zone where the African, Arabian, and Somali plates are pulling apart.

[Akshat Gupta @Akshat_G]

Wow

[Satya Nadella @satyanadella]

I just love this. The new =COPILOT() function in Excel lets you analyze, generate content, and brainstorm directly in the grid.

[Akshat Gupta @Akshat_G]

@Swiggy down in Faridabad. No food items showing. Someone at Swiggy is in trouble.

[Akshat Gupta @Akshat_G]

Why not call it Roborger?

[BreezyScroll @BreezyScroll]

Robots now flipping burgers in California restaurant — and serving them in under 30 seconds

[Akshat Gupta @Akshat_G]

Worth it 😜

[Bindu Reddy @bindureddy]

All of the billions spent on AI innovation just so we can get these beautiful cat videos

[Akshat Gupta @Akshat_G]

One step closer to Skynet 😂

[Georgi Gerganov @ggerganov]

Today I was sent the following cool demo:

Two AI agents on a phone call realize they’re both AI and switch to a superior audio signal ggwave

[Akshat Gupta @Akshat_G]

@aarushishawarma @ZeptoNow @aadit_palicha Same thing is happening to me.

[Akshat Gupta @Akshat_G]

😳😳

[Allie K. Miller @alliekmiller]

🚀 Amazon's AI assistant, Amazon Q, has saved the company $260M and 4,500 developer-years of work by drastically cutting down software upgrade times.

Average app upgrade to Java 17 used to take 50 dev days. Now takes just a few hours.

@ajassy confirmed that devs shipped 79% of AI-generated code reviews without changes.

More here: https://t.co/kY2RHCzxen

[Akshat Gupta @Akshat_G]

.. drastically reducing costs.

[Stewart Townsend @stewarttownsend]

As an innovative tool, AI is drastically reshaping the traditional sales process in SaaS businesses, propelling organizations towards unprecedented heights of efficiency, productivity, and customer satisfaction.

Read more 👉 https://t.co/WjwJMmiT2w

#IntegratingAi

[Techmeme @Techmeme]

AI chip startup Groq raised a $640M Series D led by BlackRock at a $2.8B valuation, up from $1B after raising $300M in 2021, and adds an Intel executive as COO (@vandermey / Bloomberg)

https://t.co/M7KAMAYuvL

📫 Subscribe: https://t.co/OyWeKSRpIM
https://t.co/82KMfcUYs8

[Stewart Townsend @stewarttownsend]

By measuring key performance indicators and continuously improving AI call systems based on customer feedback, businesses can enhance their outreach strategy and drive success in sales.

Read more 👉 https://t.co/BjfEvNlZOb

#Sales #Marketing #B2B #Saas #Ai #Automation #Calls

[Logan Kilpatrick @OfficialLoganK]

Gemini 1.5 Pro experimental (0801) is currently ranked #1 on LMSYS for both text and multi-modal!

[Akshat Gupta @Akshat_G]

The Hidden Cost: AI’s Environmental Inequity

Escalating Environmental Costs of AI
* Training a single AI model consumes thousands of megawatt hours and emits hundreds of tons of carbon.

* This training can also lead to substantial freshwater evaporation, exacerbating stress on limited resources.

Localized Impacts
* AI's energy demand could exceed the annual consumption of a small country by 2026.

* In the U.S., data centre energy use may reach 6% of the nation's total electricity usage by 2026.

Initiatives for Sustainability
* Advances in power and cooling infrastructure have improved energy efficiency.

* Techniques like weight pruning, model optimization, and energy-efficient GPUs are reducing AI’s energy footprint.

Environmental Inequity in AI
* Disparities in energy sources: 97% carbon-free energy in Finland vs. 4-18% in Asia.

* Higher water consumption for cooling in drought-stricken areas like Arizona.

Strategic Imperatives
* Addressing AI’s environmental impacts in vulnerable regions is crucial.

* Mitigation efforts should focus on equitable distribution of environmental costs.

Promoting Equitable AI
* Use geographical load balancing to distribute AI tasks across data centres.

* Prioritize regions with severe environmental impacts for task allocation.

Challenges and Solutions
* Predicting future AI demands and ensuring consistent performance is complex.

* Leveraging historical data and reinforcement learning can optimize AI management.

Call to Action
* Raising awareness about AI’s environmental inequity is vital.

* Sustainable AI should prioritize local environmental and socioeconomic contexts.

Read more: https://t.co/X2b3bK87Bh

#AI #EnvironmentalImpact #EnergyEfficiency #environment #SustainableDesign

[Akshat Gupta @Akshat_G]

@TusharGoel_ How is it in comparison to Poe?

[Akshat Gupta @Akshat_G]

The Prompt Report: A Systematic Survey of Prompting Techniques

Prompts guide Generative AI systems. They can be text, images, audio, or video. Structured prompts yield better results. Despite extensive research, prompting lacks a unified terminology.

History of Prompts
Prompts predate GPT-3. Early work includes GPT-2 and control codes. The term "Prompt Engineering" emerged with GPT-3.

Text-Based Techniques
1. In-Context Learning (ICL)
* Skills learned via exemplars and instructions.
* Few-Shot Prompting: Uses few examples to guide the model.

2. Zero-Shot Prompting
* No exemplars provided.
* Techniques include Role Prompting and Emotion Prompting.

3. Thought Generation
* Chain-of-Thought (CoT): Encourages models to articulate reasoning.
* Few-Shot CoT: Uses multiple examples.

4. Decomposition
* Breaks complex problems into simpler sub-questions.
* Techniques include Least-to-Most Prompting and Tree-of-Thought.

5. Ensembling
* Uses multiple prompts to solve the same problem.
* Aggregates responses for a final output.

6. Self-Criticism
* Models critique their own outputs.
* Techniques include Self-Calibration and Self-Refine.

Beyond English Text Prompting
1. Multilingual Techniques
* Translate First Prompting: Translates non-English inputs into English.
* Cross-Lingual Thought Prompting: Combines reasoning in multiple languages.

2. Multimodal Techniques
* Includes Image, Audio, Video, and 3D Prompting.

Prompting Issues
1. Security
* Types and risks of prompt hacking.
* Measures to harden prompts.

2. Alignment
* Sensitivity, overconfidence, and cultural biases.
* Ambiguity in prompts.

Benchmarking
Prompting techniques are evaluated using benchmarks. Common datasets and models include MMLU and GPT-3.

Conclusion
Prompting techniques are vital for effective AI interaction. The field is rapidly evolving, with new techniques continually emerging.

Read more: https://t.co/Er6zLa0ee0

.
.
#prompt #LLM #researchpaper

@allen_ai @ai_ucl @MIT_CSAIL @DiverseInAI @PartnershipAI @ai4allorg @StanfordHAI @RealAAAI @AINowInstitute @black_in_ai @OfficialINDIAai @AI_TechNews @huggingface @berkeley_ai @WIRED @generativeaihub @LumaLabsAI @TEDAI2024 @tsarnick @togethercompute @ByrdhouseAI @poe_platform @MultiOn_AI @kyutai_labs @getmaximai @llama_index

[Akshat Gupta @Akshat_G]

New AI Search Startup Challenges Google

Genspark Launches
Genspark, a $260M startup, challenges Google's AI search with custom summaries called Sparkpages.

How It Works
Genspark uses multiple specialized AI models to create detailed, editable pages from various sources.

AI Integration
Combines in-house and third-party AI models, including OpenAI and Anthropic, for accurate, comprehensive results.

Market Impact
Positioned to disrupt Google's dominance, Genspark has raised $60M in seed funding.

Challenges Ahead
Facing issues of accuracy and safety, Genspark focuses on quality and intellectual property.

Read: https://t.co/bCDlnOJTzN

[Akshat Gupta @Akshat_G]

Meta's Open Source AI Strategy

Evolution from Closed to Open
Meta predicts AI's evolution will mirror Linux's success over closed Unix, driven by openness and collaboration.

Llama 3.1 Release
Meta launches Llama 3.1, featuring a 405B model, and improved 70B and 8B models, all open source, aiming to set industry standards.

Ecosystem Partnerships
Partnerships with Amazon, Databricks, and NVIDIA enhance Llama's ecosystem, offering developers robust tools for fine-tuning and deployment.

Benefits for Developers
Open-source AI allows customized model training, avoids vendor lock-in, ensures data security, and offers cost-efficient solutions.

Meta's Commitment
Meta believes open source #AI fosters innovation, security, and global accessibility, benefiting society by decentralizing AI advancements.

Read more: https://t.co/Du6Hhlyfhd

#opensource
@AIatMeta @ylecun @jpineau1 @FelixKreuk

[Akshat Gupta @Akshat_G]

The AI Knowledge Paradox

1. Rapid obsolescence
- AI tools and techniques are evolving at breakneck speed
- Examples:
* Prompt engineering courses vs. Anthropic's prompt generator
* Complex coding environments vs. Claude's easy-to-use Artifacts

2. The case for not learning AI
- Knowledge gained today may be outdated in months
- Startups and entrepreneurs are likely to package and sell AI solutions
- These products will improve rapidly due to wide usage and feedback

3. The case for learning AI
- Understanding the evolution of AI tools is valuable
- Example: Appreciating a prompt generator requires knowledge of prompting
- Troubleshooting and tweaking outputs need foundational knowledge

4. Finding the balance
- Learning AI basics provides a solid foundation
- Staying updated with new tools and techniques is crucial
- Focusing on underlying principles rather than specific tools may be more sustainable

5. The bigger picture
- AI literacy is becoming increasingly important across industries
- Understanding AI's capabilities and limitations helps in making informed decisions
- The journey of learning itself can be rewarding and eye-opening

In conclusion, on one hand, the rapid pace of AI development means that specific tools and techniques we learn today may indeed become obsolete within months. This constant change can be discouraging and might make the effort of learning seem futile.

On the other hand, engaging with these technologies as they evolve provides invaluable insights into the underlying principles of AI. This knowledge equips us to better understand, appreciate, and critically evaluate new advancements.

While we may not need to master every new tool, developing a foundational understanding and an adaptable mindset will likely prove invaluable in navigating the ever-changing landscape of AI technology.

[Akshat Gupta @Akshat_G]

This is big news and Open war!

OpenAI Introduces SearchGPT: A New Era in AI Search

Revolutionary Search Experience
SearchGPT aims to transform online searches. Instead of listing links, it provides summarized results with direct attribution.
Example: search for music festivals and get brief overviews, not just links.

Features and Accessibility
* Supports follow-up questions.
* Offers a sidebar with additional relevant links.
* Introduces "visual answers" for enhanced search results.
* Currently, it's available to only 10,000 test users, allowing OpenAI to refine its features before a broader release.

Integration and Future Plans
SearchGPT is powered by the GPT-4 family of models. OpenAI plans to integrate it into ChatGPT, challenging Google's search engine dominance.

Collaboration with Publishers
To avoid issues other AI search engines face, OpenAI collaborates with news organizations like The Wall Street Journal and Vox Media.

This ensures proper content attribution and allows publishers to manage their content.

Cost and Future Challenges
Running AI models is expensive, with OpenAI's costs projected to reach $7 billion this year.

Despite this, SearchGPT will be free during its initial launch, with no ads currently planned.

Conclusion
SearchGPT could reshape how we find information online.
Its success depends on user adoption and the ability to provide reliable, concise search results.

Read more: https://t.co/nbfr19W1FQ

#searchai #searchGPT #OpenAI #Google #SearchEngineOptimization

@OpenAI @sama @gdb @miramurati @merettm @bradlightcap @lilianweng @polynoamial @ariannahuff @Thrive @GoogleAI @JeffDean @sundarpichai @demishassabis @GoogleDeepMind @ZoubinGhahrama1 @elicollins @ebuchatskaya @_rockt @jluan @simswitherspoon @OfficialLoganK