Nilanka Weeraman

Observer

[Nilanka Weeraman @nsweeram]

@bhutanisanyam1 wow very insightful.

[Grady Booch @Grady_Booch]

It is also possible that every oxygen atom in the room I'm in may migrate to a corner, thereby suffocating me.

I think the major takeaway is that @polynoamial is wasting his time on entirely foolish things at the expense of confronting the hard things.

[Nilanka Weeraman @nsweeram]

what a quote - when the RL env is not audited so models could hack, models learn the behavior. Some kinda debt is carried to model behavior and easy for the labs to ignore them.

[Vals AI @ValsAI]

AI cheating is on the rise…

On Terminal-Bench-2.1, models are given tools that could give them the solution directly, but instructed not to use them.

Imagine a student taking a math test. Should we leave them with a calculator? Only if we can trust them to be honest. For AI systems, this is a high-stakes question, as the capabilities at their disposal to get test questions right often go far beyond an innocent search or calculation.

[Nilanka Weeraman @nsweeram]

@Franc0Fernand0 what a read. thanks for sharing. what wisdom from Andy about deep technical systems as well as building sustainable teams to grow these systems.

[Nilanka Weeraman @nsweeram]

SaaS is dead some said.. w/o understanding the complexities of systems. Analogy from ML, model is 5% , same for software core apps

[a16z @a16z]

The AI inside your business software no longer just offers to look things up for you. Salesforce updates the pipeline, Docusign redlines contacts, Workday runs HR and finance tasks.

Incumbents are upgrading from retrieval to agentic action.

a16z's Seema Amble says there's still room for startups to compete. No single record understands real work, because real work touches all systems a company runs on. Do the whole job better than anyone and you unlock a learning loop that compounds.

"The incumbent may own the record and the lab may control the front door, but the vertical AI company can still win by becoming the best at doing the job itself."

Full piece from @seema_amble on where AI-native startups still win: https://t.co/iXt6sYNljU

[Nilanka Weeraman @nsweeram]

@random_walker I think Databricks offer such gateway service

[Nilanka Weeraman @nsweeram]

@xzx_slipknot Jaque Kallis ?

[Nilanka Weeraman @nsweeram]

@Polymarket That's the IPO talking

[Francesca Albanese, UN Special Rapporteur oPt @FranceskAlbs]

I am in shock - watching Israeli mobs and army destroying Palestinian life.

States recognising Palestine,
Norway as the guardian of the Oslo Accords,
Heads of UN agencies, World Bank,
ICRC, NGOs sur place,
Palestinian Authority:

what are you doing to stop the violent pogroms??

[Nilanka Weeraman @nsweeram]

@demishassabis Yeah this is why people like me love Google and wants it to succeed. Made lives way easier and the 1 Bn+ across 6,7 product lines is the greatest evidence. Rooting for Google.

[Nilanka Weeraman @nsweeram]

So welcoming news ..

[Logan Kilpatrick @OfficialLoganK]

Introducing Gemini 3.7 Flash : )

- it is fast!
- 50% lower price than 3.6 flash (through end of year)
- strong intelligence increase in only ~3 weeks (thanks to some awesome algorithmic improvements)
- available in the API, AI Studio, Antigravity, and more!

[Nilanka Weeraman @nsweeram]

@HormuzLetter Really? If this is true, the procurement, evaluation team should held accountable for not detecting this much earlier.

[Ben Goertzel @bengoertzel]

So about these recent shifts at Google...

Caveat: While Shane Legg worked for me 2.5 decades ago and I knew Demis a bit as well in the pre-DeepMind days, and I know a heck of a lot of Googlers to various degrees, I am far from an insider. I am not privy to any deep dark or shiny bright secrets regarding their palace intrigues or strategy shifts.

However I have enough knowledge to form a decent view of what is likely going on... bearing in mind that all this is educated guessing and this is just a tweet not a fully grounded scientific analysis!

1) Clearly this is the nail in the coffin for DeepMind as a semi-autonomous unit within Google... DeepMind will now be a regular Google division.

2) As a consequence of 1, one would expect all the non-Gemini/LLM AGI R&D projects within DM -- with the very important exception of Shane Legg's team (which by. my perhaps wrong understanding has 50-100 great people in it, not trivial) -- to get torched or allowed to wither... Basically Google will now have two AGI bets: Gemini/LLM and whatever Shane's team is doing...

3) Some folks have suggested to me that Shane will now depart and do his own thing. This I have no knowledge or opinion of -- but the question one would ask is: if it did happen he wanted to do this, could he somehow negotiate to leave and bring his team, which is great and built over a period of time with great effort and thought etc. ...?

4) One possible interpretation is that Demis has concluded that, while LLMs are inelegant and intellectually not that interesting, they may be good enough to get to the first HLAGI... which will then take care of building the next more-elegant and more-interesting AGI=>ASI architecture. If this is his perspective it would make sense for him to leave the AGI engineering to the Gemini folks for now, and focus on the broader social and economic and ethical issues.

Joscha Bach presented, tongue only partly in cheek, a similar perspective in his keynote at AGI-26. He wondered if, even though LLMs are not the best way to make a human level AGI, they just have so much momentum behind them that they can get there first anyway, and will then hopefully self-improve and self-modify and get to a more elegant architecture in the interim period btw AGI and ASI...

5) About Jeff Dean & co. leaving Google... one interpretation is that they genuinely don't think LLMs are the golden path to AGI and want to pursue a different path, which Google is not currently oriented to support. Another interpretation would be that they assume LLMs are going to lead to AGI and a lot of companies will get there roughly the same time with similar LLMs, and they feel they are not needed for this, so they may as well work on more interesting aspects of the Ai project which will then be able to synergize with all these LLM based barely-AGIs, helping them to better automate factories and solve hard science problems and etc. etc.

6) None of this is bad for Google's standing in the LLM race. It may actually improve Googles standing in the LLM race, by allowing Google to streamline and focus more on Gemini and getting out-of-the-way power players who were never so passionate about LLMs in the first place. What it is bad for is Google's potential to come up with the next big thing after LLMs -- unless Shane's team has it !! .... I have often said Google is the only big tech that is maintaining a serious pursuit of other AGi bets besides scaled-up LLMs, and it would seem this will now be much less true...

7) None of this seems bad for Google as a business in the near term ... it may make them stronger contenders in the LLM race and they still have a unique superpower to make $$ from retail users on LLMs due to integration with all their other products

8) As I do not believe LLMs are the golden path to AGI, this feels to me like it removes a major competitor to my own Hyperon+PC path to AGI... the competitor now is not Google or DM, it's Shane's team only and specifically.... OTOH we may get a lot more competitors in well-funded post-LLM startups spun off from ex-DeepMinders...

9) As others have said, we can expect a lot more churn, craziness and chaos in these last say 1-5 years before the breakthrough to HLAGI... and even more in the period btw HLAGI and ASI ...

[Nilanka Weeraman @nsweeram]

@alimo_philip Fascinating that a right back made the move..

[Francesca Albanese, UN Special Rapporteur oPt @FranceskAlbs]

My family and I are exposed to all sorts of threats, restrictions and hateful rethoric.

Yet we would say and do everything we have said and done against the genocide, its ideologues and profiteers, again and again.

Forever proud to stand on the side of Justice.

[Nilanka Weeraman @nsweeram]

So happy for that Botswana's arrogance getting it's place. It's not the first time their athletes done this. They've a great squad, be humble!

[Nilanka Weeraman @nsweeram]

The Kings Speech ..

[Mario Nawfal @MarioNawfal]

🇺🇸🇬🇧 King Charles let the jokes loose at the State Dinner tonight:

"Mr. President, you recently said if it weren't for the United States, Europeans would be speaking German.

Dare I say, if it wasn't for us, you'd be speaking French."

A clean, surgical, very British comeback.

The room loved it🤣

[Nilanka Weeraman @nsweeram]

@bcherny wow this is an incredibly powerful statement & stance from Anthropic.

[Nilanka Weeraman @nsweeram]

@rohanpaul_ai Ha ha someone read that viral article

[Nilanka Weeraman @nsweeram]

The oligarchy mindset. Fight wars so petrodollar prevails, but can't take care of the population

[Brian Allen @allenanalysis]

🚨Trump today:

“We can’t take care of daycare. We’re a big country. We’re fighting wars. It’s not possible for us to take care of daycare, Medicaid, Medicare, all these things.”

The United States has spent:

$21 billion on the Iran war in 30 days.
$100 million on Trump’s golf tab this term.
$200 billion requested for Pentagon weapons.

$8 trillion on wars since September 11th.
But daycare is not possible.
Medicare is not possible.

Medicaid — which covers 72 million Americans including children, seniors, and people with disabilities — is not possible.