Python Dev, Parent. Author - https://xitter.twitterapi.xyz/t.co/5lts7q9z7R Curator - https://xitter.twitterapi.xyz/t.co/wr74oHNs8O Creator - MapToPoster https://xitter.twitterapi.xyz/t.co/YQt2CoiupJ 🖖
Planet Earth
Joined January 2012
- Tweets7K
- Following3.4K
- Followers36.7K
- Likes14.2K
During an Agentic AI long running plan, micro management be like
Give me status every 5 minutes
6
412
Ankur Gupta retweeted
Amazon and AWS switched their internal ticketing label system to a decision model, away from Sonnet. It's 4.5x faster and 44x cheaper. Same accuracy. strandsagents.com/blog/amazo…
The philosophical tragedy of modern life is that the algorithm knows our subconscious desires better than ourselves, simply because we won't stop providing free reinforcement learning at 2 AM.
Ankur Gupta retweeted
Announcing a major milestone achieved by my team @GoogleDeepMind: we have proven key security properties of the machine code underlying the Signal messaging protocol. This is the first time the actual production code of a widely deployed protocol has been formally verified – by any method, with or without AI – and at this depth.
More than a research breakthrough, it’s a prime example of how frontier AI can be a force multiplier to counter the speed and scale of modern cyber threats.
Formal verification offers a higher standard of security; however, it’s extremely complex and time consuming, which is why in many cases it’s considered to be impractical. Similar efforts could take teams decades of combined effort. We did it in a month’s time.
Using AlphaProof Code, Google DeepMind's new agentic framework for formal verification that runs on LLMs including Gemini Carbon, we verified the Signal protocol down to the actual code running on devices. This milestone lays the groundwork for a future in which vital systems, from medical device firmware to power grid software, ship with mathematical proof that their protections work.
Reaching that future will take continued research from our community. In that spirit, we're open-sourcing our formal specifications and verification artifacts so others can build on this work.
On a personal note, this moment is of deep significance to me. I started my own research journey more than two decades ago as an intern working on checking the consistency of complex software systems. Seeing this grand challenge made practical with modern AI is extraordinary.
I'm extremely proud of my team members: Swarat Chaudhuri, Andrew Ferraiuolo, Anton Kovsharov, Tej Chajed, Sam Schlesinger, Karn Seth, Jonathan Katz, and Charles Sutton for taking on this grand challenge and delivering, and grateful to the Signal team for their partnership.
deepmind.google/blog/formal-…
Ankur Gupta retweeted
DeepSeek's DSH harness is built on a simple idea: everything is a plugin. Including the agent loop itself.
In this tutorial I break down cordis, the plugin system behind it:
> the architecture of dsh
> services, dependency injection, and events
> building a real plugin live
Thread 👇
During Diwali as one would expect, you connect with friends, family, ex-colleagues etc to wish and have short conversations.
Man pretty much every second conversation is about AI and it's impact on job market, specially when most of the people I know are in Tech or their kids are.
Sort of a spin-off thought from the conversations was that we Indians should engage with financial planner for advice and review on long term financial goals/outcomes. Just review and advise, I feel we can get better returns and achieve them faster.
10
1,153
Shubh Deepavali! 🪔
Wishing everyone strength of body, peace of mind, and joy in every celebration.
8
1,191
Ankur Gupta retweeted
. @hdfcmf set to launch a charity fund for cancer cure. The special fixed maturity plan will open on November 16 and close on Nov 30.
The 1169-day close-ended debt scheme offers principal redemption on maturity. Investors can donate 50% or 75% of income to Indian Cancer Society, directly. @IndianCancerSoc
In past, 17,453 underprivileged cancer patients have been funded through previous editions of this fund -- Rs273 cr across 22 hosptials
Ankur Gupta retweeted
Paper Breakdown - Harness2 Recursive Agent Harnessing for an Open World
Harness2 shows that changing an agent’s workflow on today’s task, then retain evidence about those improvements to make future workflow refinement. Raises GLM-5’s WorkBuddy-Bench average from 64.2 to 75.7 🧵
Ankur Gupta retweeted
Theory says sell outcomes. But in practice try getting procurement to approve the invoice.
Outcome pricing needs agreement on what changed, who gets credit, and what finance will pay for it. The buyers operation team has to be ready for that contract too.
“Breaches prevented” sounds great until someone asks you to prove how many.
Measure outcomes. Bill for what both sides can verify. That often means outputs, even in AI.
Are you measuring outcomes or outputs?
Vanta CEO Christina Cacioppo (@christinacaci) said it right in a recent interview. Every AI company claims to price on outcomes, but most of what gets called an outcome is actually an output: documents drafted, alert closed, account offboard.
An output is something the software did. An outcome is something that changed for the customer because of it.
How they are different:
-- A drafted contract is an output. A deal that closed faster is an outcome.
-- A closed alert is an output. A breach that didn’t happen is an outcome.
-- An offboarded account is an output. Zero lingering access when the auditor shows up is the outcome.
Outputs are easy to count, which is exactly why AI startups bill based on outputs. But billing on outputs rewards volume, not value. Pay per alert closed, and you’ve built a system that’s rewarded for closing alerts, not for making you more secure.
AI Buyers: ask what you’d measure six months from now to know it worked. If the vendor’s pricing unit isn’t tied to that, you’re paying for activity.
AI Startups: real outcome pricing means taking on risk you don’t fully control. That’s hard. Calling per-task pricing “outcome-based” is easy, and customers are starting to notice the difference.
This goes to the recent discussion around CONSEQUENTIAL and VERIFIABLE. Buyers value (and pay the most for) consequential, non-verifiable (long horizon) outcomes.
Outputs are what the software does. Outcomes are why anyone buys it.
If token speed were to 10x from here and be available for all at affordable price. Many software features in an existing software be on-demand in the request-response cycle itself such that you describe the extension or additional feature you need and get the corresponding code generated at runtime and plugged into the platform and you can simply pin it for your usage post trying it out.
Such behaviour or pattern might be half a decade away but it's coming once every org has a software factory building all its software and compute needed for same due to smaller more efficient models from inference standpoint are widely available from open source community.
5
1,008
Ankur Gupta retweeted
Worth a read if you work with spatial data: a piece from @GISGeography on some geospatial concepts - 𝐇𝐨𝐰 𝐆𝐈𝐒 𝐔𝐬𝐞𝐬 𝐌𝐚𝐩𝐩𝐢𝐧𝐠 𝐎𝐯𝐞𝐫𝐥𝐚𝐲𝐬
More: gisgeography.com/mapping-ove…
𝐓𝐡𝐞 𝐍𝐞𝐰 𝐒𝐜𝐢𝐞𝐧𝐜𝐞 𝐨𝐟 𝐌𝐚𝐩𝐬 - thenewscienceofmaps.com/
