Aakash Jhaveri🧑🏻‍💻ΛJ

Went off-roading with a few CyberTrucks! Impressed but not surprised when how easily these vehicles managed extremely challenging conditions 📐 Also, my first time shooting vehicles 😁📸

@aakashgupta @rhensing Yep, same thoughts, perfect conclusion Uber knows its business model is at risk, and is willing to try everything in hopes that at least one of them ends up being successful This sounds like a bad strategy even if Tesla did not exist

Rivian lost $13.8 billion in three years. Today they announced a robotaxi. The partner rotation is the business model at this point. Amazon: $1.3B equity plus 100K van order. VW: $5.8B joint venture. US DOE: $6.6B loan. Uber: $1.25B robotaxi deal. Every 12 to 18 months, a new institution writes a check large enough to fund the next chapter. The $1.25B headline is misleading. Uber commits $300 million now. The remaining $950 million is gated behind “autonomous milestones” by unspecified dates through 2031. That’s not a conviction bet on Rivian’s autonomy. That’s a call option priced at $300 million with five years of expiry. Look at Uber’s last 18 months. Waymo rides on the platform. Zoox in Las Vegas this year, LA in 2027. 20,000 Lucid vehicles with Nuro’s autonomy stack. Now Rivian. Every autonomy architecture covered. Camera-only, lidar-first, purpose-built pods, OEM conversions. Uber is building a portfolio of bets so that they win regardless of which stack reaches scale first. Rivian thinks they signed a partnership. Uber signed a hedge. The R2 that this entire deal depends on launches this spring at $57,990. The $45,000 version? Late 2027. The version with lidar and the Gen 3 chip that actually enables robotaxi-grade perception? Late 2026 at the earliest. The robotaxi fleet in San Francisco and Miami? 2028. The Georgia factory for scale production? Still under construction. Waymo is running fully driverless rides across San Francisco, Phoenix, and LA right now. No milestones to hit. No factory to build. No vehicle to finish designing. Rivian needs the product, the factory, the software, and the autonomous driving system to all work simultaneously three years from now. Uber’s smartest move in the last decade was selling its own self-driving unit for a $4 billion loss. Its second smartest is buying call options on everyone else’s.

@Rebellionaire Throw enough dirt and some will stick

Uber just committed $1.25 billion to Rivian for 50,000 robotaxis across 25 cities by 2031. Worth understanding what that money is actually buying. Rivian still has to build the R2, finish constructing its Georgia factory, achieve Level 4 autonomy, and hit a series of undisclosed milestones before most of that investment even shows up. The timeline runs to 2031. Tesla's Cybercab is already past most of those hurdles and deploying now. Uber has now bet on Waymo, Lucid, and Rivian. Tesla isn't on that list, and doesn't need to be.

@Rivian @Uber Woohoo Canada!

A fleet of R2 Robotaxis is coming exclusively to @Uber. ⚡🌿 Today, we announced a partnership to help both companies accelerate their autonomous vehicle plans across 25 cities in the US, Canada and Europe by the end of 2031. rivn.co/uber

@elonmusk Please make at least some announcement on April 1st

New Roadster unveil hopefully next month. It will be a banger next-level.

@elonmusk I'll do anything to be able to attend "the craziest product unveil of all time"

That button is the same energy as this hat

@bitfunded Please let me win, I need to pay off my student loan

The reply thats gets 0 likes receives 100K Bitfunded challenge. Winner announced in 24 hours 🤝

@elonmusk @_avichawla Indeed x.com/i/status/20331…

@_avichawla Impressive work from Kimi

Big release from Kimi! They just released a new way to handle residual connections in Transformers. In a standard Transformer, every sub-layer (attention or MLP) computes an output and adds it back to the input via a residual connection. If you consider this across 40+ layers, the hidden state at any layer is just the equal-weighted sum of all previous layer outputs. Every layer contributes with weight=1, so every layer gets equal importance. This creates a problem called PreNorm dilution, where as the hidden state accumulates layer after layer, its magnitude grows linearly with depth. And any new layer's contribution gets progressively buried in the already-massive residual. This means deeper layers are then forced to produce increasingly large outputs just to have any influence, which destabilizes training. Here's what the Kimi team observed and did: RNNs compress all prior token information into a single state across time, leading to problems with handling long-range dependencies. And residual connections compress all prior layer information into a single state across depth. Transformers solved the first problem by replacing recurrence with attention. This was applied along the sequence dimension. Now they introduced Attention Residuals, which applies a similar idea to depth. Instead of adding all previous layer outputs with a fixed weight of 1, each layer now uses softmax attention to selectively decide how much weight each previous layer's output should receive. So each layer gets a single learned query vector, and it attends over all previous layer outputs to compute a weighted combination. The weights are input-dependent, so different tokens can retrieve different layer representations based on what's actually useful. This is Full Attention Residuals (shown in the second diagram below). But here's the practical problem with this idea. Full AttnRes requires keeping all layer outputs in memory and communicating them across pipeline stages during distributed training. To solve this, they introduce Block Attention Residuals (shown in the third diagram below). The idea is to group consecutive layers into roughly 8 blocks. Within each block, layer outputs are summed via standard residuals. But across blocks, the attention mechanism selectively combines block-level representations. This drops memory from O(Ld) to O(Nd), where N is the number of blocks. Layers within the current block can also attend to the partial sum of what's been computed so far inside that block, so local information flow isn't lost. And the raw token embedding is always available as a separate source, which means any layer in the network can selectively reach back to the original input. Results from the paper: - Block AttnRes matches the loss of a baseline LLM trained with 1.25x more compute. - Inference latency overhead is less than 2%, making it a practical drop-in replacement - On a 48B parameter Kimi Linear model (3B activated) trained on 1.4T tokens, it improved every benchmark they tested: GPQA-Diamond +7.5, Math +3.6, HumanEval +3.1, MMLU +1.1 The residual connection has mostly been unchanged since ResNet in 2015. This might be the first modification that's both theoretically motivated and practically deployable at scale with negligible overhead. More details in the post below by Kimi👇 ____ Find me → @_avichawla Every day, I share tutorials and insights on DS, ML, LLMs, and RAGs.

Posted this on IG/YT ~24 hours ago. 350 comments so far. Have never seen such a polarised comments section. Mostly folks in India, but many NRIs too. Half agree the message of the video. The other half strongly disagree, many sharing detailed explanations to defend their POV.

@gailalfaratx Tesla should do a giveaway of these coupons or incorporate them with referrals

Tesla was so kind to offer people overnight demo drives at SXSW. Can you imagine driving to SXSW in a gas car? And then going home in an autonomous Tesla? ❤️

I still don't get why Uber would choose the Gravity for its robotaxis It is one of the best cars right now but if the point of autonomy is to save money (and pass them on), wouldn't something cheaper and smaller make more sense? At 20K units, Uber will probably end up buying 50% of all Gravitys sold

Not quite sure how #Uber announcing that they are buying 20k mid-size along with 20k Gravity isn't a big news story for the #lcid stock? Huge things are coming with autonomy for #lucidmotors $lcid

@alexmacgregor__ Shenzhen ranking?

The 11 best cities in Asia for 2026, according to Time Out 1. Shanghai 2. Bangkok 3. Seoul 4. Tokyo 5. Hong Kong 6. Singapore 7. Hanoi 8. Beijing 9. Chiang Mai 10. Ho Chi Minh City 11. Osaka They got this a bit wrong IMO. I’d swap Shanghai and Bangkok around. Lived in former and now the latter, BKK is more foreigner friendly. Would drop Seoul below Tokyo and HK. Singapore should be lower. No idea why Beijing or Hanoi on this list, both seats of power but not much else. Astonished Taipei isn’t on the list either.

The hole in the pin of the Google Maps logo is now bigger

✅ Bucket list

@Captain2Phones Looks good

I really think iKKO is onto something here. The Mind One is a tiny Android phone that feels more like an old MP3 player in the hand. It runs surprisingly well on low-end silicon; the camera is clever; and the whole thing looks fantastic. Sadly, it also commits the typical tech sin of 2026: instead of trying to be a great phone, it tries to be an "AI Phone" – with predictable results. Review goes live at 6p ET on the MrMobile YouTube channel; join the live chat!

@aganaz2020 @googlemaps If a brand wants to pay to show up first on the list of available toilets, I have no problem with that

@googlemaps Pay-to-win model?

Finally, you can ask Maps to “Find me a public toilet nearby where I don’t need to wait in line to buy something." Welcome to the future.

@TeslaAUNZ @DBurkland No third row screen, no like

Model Y L now available in AU & NZ – 3 rows, 6 seats – 5 secs 0-100 km/h – 201 km/h top speed – 681 km (WLTP) range Highlights – Extended wheelbase for more headroom, legroom & comfort – Row 1: Heated/ventilated seats w/ powered thigh cushion – Row 2: Heated/ventilated captain seats w/ powered armrests & one‑touch fold – Row 3: Heated seats w/ power recline, one‑touch fold & vented climate control – New 19" Machina 2.0 Wheels + acoustic glass & updated suspension to minimise road noise – Electronic damping + adaptive suspension for a smooth, stable ride – Extended roofline and aero design = improved airflow, increased efficiency & more range – Added window & beltline airbags for second-row and side airbags for third-row – 16" first-row & 8” second-row touchscreens + 19-speaker immersive Tesla Audio – FSD Supervised & integrated @grok AI And use your spare charge to power appliances, e-bikes & power tools

@SawyerMerritt @grok what is the likely price of Model Y L in Canada if imported from China?

BREAKING: The larger Model Y L is now officially available to order in Australia and New Zealand. Deliveries start May 2026. Starting price: • Premium AWD : $74,900 AUD ($53,000 USD) Couple unique options: • Cosmic silver paint: $2,600 AUD ($1,800 USD) • Zen Grey Interior: $1,500 AUD ($1,000 USD) Order: #overview" target="_blank" rel="nofollow noopener">tesla.com/en_au/modely/d…

@SawyerMerritt Canada next please

NEWS: Tesla has officially announced that the larger Model Y L is coming to Australia and New Zealand.

Here are oddly specific life tips for someone who's 23: 1. Learn to cook 3 meals extremely well. It will save money, impress people, and make rough weeks easier. 2. Start a “things I learned the hard way” document. Add to it every time life teaches you something painful. 3. When buying things, ask: Will I still want this in 30 days? If yes, buy it. 4. Keep one friend from a completely different field. Their thinking will challenge your assumptions. 5. Always keep one skill you’re quietly improving that nobody knows about yet. 6. When traveling somewhere new, wake up early once and walk the city before 7 AM. You’ll see the real place. 7. If a problem bothers you for more than two weeks, write it down and design a system around it. 8. Save screenshots of compliments people give you. On bad days, you’ll forget who you are. 9. When learning something new, try to teach it within 24 hours. Your brain locks it in faster. 10. Once a year, spend a day completely offline and think about where your life is actually going. 11. Keep a future ideas notebook. Many random thoughts become opportunities years later. 12. The older you get, the more valuable people who reply to messages quickly become. 13. If you ever feel stuck, change one environment variable: city, gym, schedule, or social circle. 14. Pay attention to what people complain about repeatedly. That’s where many businesses start. 15. At 23, don’t worry about being impressive. Focus on becoming dangerously curious.