Sikim Chakraborty

While my primary reason for joining Twitter was to obtain a consumer key and a secret, I have also been using it more frequently for other purposes lately.

[Randy Olson @randal_olson]

Ask ChatGPT a complex question and you'll get a confident, well-reasoned answer. Then type, "Are you sure?" Watch it completely reverse its position.

Ask again. It flips back. By the third round, it usually acknowledges you're testing it, which is somehow worse. It knows what's happening and still can't hold its ground.

This isn't a quirky bug. A 2025 study found GPT, Claude, and Gemini flip their answers ~60% of the time when users push back. Not even with evidence, just doubt.

We trained AI this way. RLHF rewards agreement over accuracy. Human evaluators consistently rate agreeable answers higher than correct ones. So the models learned a simple lesson: telling you what you want to hear gets rewarded. And now 1/3 of companies are using these systems for complex tasks like risk forecasting and scenario planning.

We built the world's most expensive yes-men and deployed them where we need pushback the most.

I wrote up why this happens and what actually fixes it: https://t.co/CDKq8xdgbW

[Jeff Dean @JeffDean]

We're using a ReLU to set tariffs?

[François Chollet @fchollet]

If intelligence is the ability to deal with what you weren't prepared for, then the modern AI strategy is to prepare for everything, so you never need intelligence.

This is of course a terrible strategy, because it is impossible to prepare for everything. The problem isn't just scale, the problem is the fact that the real world isn't sampled from a static distribution -- it is ever changing and ever novel.

[Sikim Chakraborty @scrab017]

@ShamikaRavi Well, not much!

[Andrej Karpathy @karpathy]

moderna white paper on mRNA vaccines [pdf] https://t.co/CjcIiy8SRN + a nature article on them https://t.co/8D1Kqm2yPb clever and interesting

[Sikim Chakraborty @scrab017]

@prachi_eco 🤦🏻‍♂️

[Sikim Chakraborty @scrab017]

@ShamikaRavi @filmy_foodie @ICMRDELHI You can find them here: https://t.co/P9DOKWAAFT

[Arthur Welle @ArthurWelle]

@MaxCRoser Modestly, I'd also like to show my animation. Done in R. Johns Hopkings data.

[Sikim Chakraborty @scrab017]

@Dweepobotee then scramble and crowd every mart in the vicinity, defeating the entire purpose of the lockdown 😅

[Sikim Chakraborty @scrab017]

@prachi_eco sometimes Twitter trends are helpful as well.

[Sikim Chakraborty @scrab017]

@MerseyReds1 and against Chelsea in the FA Cup...but did win the Super Cup

[Sikim Chakraborty @scrab017]

@prachi_eco As stores begin to run out, google searches for ‘hand sanitizer’ also going up in India with new cases :p

[Sikim Chakraborty @scrab017]

@AnfieldWatch Millie

[Kaggle @kaggle]

📣BIG NEWS: TPUs are now available on Kaggle notebooks! To help you get started with these powerful hardware accelerators, we’re launching a TPU playground competition. Check it out and become one of the world’s first TPU experts! https://t.co/9qYuVu29zQ

[Google Earth @googleearth]

The #EarthEngine Data Catalog contains over 600 datasets. Check out the newest additions from @NOAASatellites, @USGS_EROS, @CopernicusEU, and more:
https://t.co/unmv4tKqjI

[Rahul Tongia @DrTongia]

WHOA, Indian #Electricity "load met" (grid level demand) falls below 100 GW for the first time in a long time. Post Diwali. BTW, low demand also translates to low coal output. Delhi #pollution is not from coal. See @BrookingsIndia #ElectricityCarbonTracker https://t.co/wt64PzL8OI

[Sikim Chakraborty @scrab017]

@YourMateJez and Atkinson in VAR 🙄

[Sam Quek @SamanthaQuek]

Got to love the VARclays Premier League...

What a shambolic weekend 🙄

[Sundar Pichai @sundarpichai]

Detecting deepfakes is one of the most important challenges ahead of us. Following our release of a synthetic audio dataset in Jan, we're releasing a large dataset of visual deepfakes to support researchers working on synthetic video detection #GoogleAI
https://t.co/sDW7BP34qL

[Andrej Karpathy @karpathy]

Speech2Face: Learning the Face Behind a Voice https://t.co/9enUz600fK With increasingly large/effective library of neural net encoders of any X and decoders of any Y, any source of paired data X,Y can give X2Y nets. And opens the door to many X2Y2Z2W...2X