Freda Duan's banner

Freda Duan

@FredaDuan • 31,223 subscribers

Investing @ Altimeter Capital. No investment advice. https://t.co/e2Ro5VeL0o

Shorts

Just finished a one-week trip to China. I've now "survived" all the major (~20) L2 self-driving and robotaxi vehicles in both the US and China. Some thoughts & observations: ▶️L2 self-driving I tested major brands like $Huawei, $Li, $NIO, $Xpeng, and $Xiaomi. Overall, they exceeded my expectations. The rides were not overly cautious and handled complex situations (yes, road conditions in China are very challenging!) quite well. Nothing compares to $Tsla's approach. I see imitation learning/end-to-end as the only effective approach for self-driving. While Chinese peers perform well on main roads, they struggle on frontage roads due to reliance on high-precision maps and rule-based methods (e.g. cars stopped in the middle of the road where there was no clear white lining). Chinese EVs' self-driving capabilities are far ahead of those from US and EU brands. I doubt any Chinese players can profit from L2 self-driving, not because it’s not useful, but because it’s hard to differentiate, and price wars dominate the market in China. Chinese consumers and regulators seem much more receptive to self-driving. Even with a 5/10 self-driving capability, cars are practically *hands-free(!)* Insurance-wise, for L3+ cars, OEMs bear responsibility for incidents, so OEMs avoid labeling cars as L3+. ▶️Robotaxi I tested major brands like $Didi, and $Bidu. I'd rate equal to $Waymo, and it's ahead of other peers. However, the same issue applies here: user experience is nearly perfect (in Yizhuang, Beijing), but expansion is the real question. Chinese robotaxi companies are very sophisticated. While the rest of the world focuses on technology, Chinese peers treat it as a product, considering unit economics, operations, mass production, etc. Interestingly, most companies expressed a preference NOT to operate fleets themselves. They aim to be asset-light and let fleet managers handle operations. Policy Support: China has a very clear approval process, driven by data (autonomous driving distance, fully driverless distance, intervention rate, passenger ratings, etc.). ▶️Chinese EVs In major cities like Beijing or Shanghai, EV adoption (green license plates vs. gas cars with blue license plates) seems to be 40%+. If 40% of cars on the road are EVs, then EV penetration (defined as the % of new car sales) must already be over 50%. In shopping malls, the ground floor is filled with EV showrooms—easily 10+ brands, many of which are unfamiliar Chinese brands. It appears almost too easy to make an electric car, which is a stark contrast to the US. $Xiaomi, for example, can achieve a 10% gross profit margin in its first year of operation, compared to $RIVN's -45%. Additionally, $Xiaomi cars are priced at 30% of $RIVN's price. It's fascinating to see how China transitioned from "couldn't make their own gas cars at all (only JVs)" to "dominating EVs globally." The government deserves credit for setting the direction and executing effectively. China now controls the entire supply chain, with $CATL holding 40% of the global market share. 🔹How did it happen? The success of the industry Incentives were set just right: the government provided incentives early on to make EVs and gas cars have comparable MSRPs, allowing consumers to choose based on functionality. This approach differs from how the IRA offers incentives... Perfectly competitive market: $TSLA was brought in, and competition was welcomed, unlike the US, which has a 100% import tax on Chinese EVs. Strategic regulations: License plate restrictions were used effectively; for example, taxis and minivans are required to be EVs. 🔹The challenges Despite the success, the industry faces challenges with low-margin companies and struggling stocks. The intense competition shows no sign of ending. Well-funded global OEMs and Chinese state-owned car companies continue to subsidize, leading to new EV brands emerging annually. The natural tendency in China is to race to the bottom. I think this ties back to China's history as the "world’s factory," where manufacturers price products at "cost plus" versus the US and developing countries, which price based on "affordability/value creation." 🔹The wow EV feature >Software features that surprised me the most: - Everything in the car can be voice-controlled. Not just simple tasks like playing music; users can adjust the height of the steering wheel and set the temperature easily. - Self-parking, which $Tsla has yet to release to all FSD users, is already a table stake in China (I'd rate the quality as 10/10). >Other fun hardware features: - Mini fridges in the car - Infotainment systems - IoT: remote access the car/home via cellphone - all connected together - Heads-up displays - UV-protected glass roofs: $Xiaomi took $Tsla's design, but the glass roof of the $Xiaomi car is made of double layers with silver, blocking 99.9% of UV and infrared rays...as a result, heat is no longer a problem inside the car

Just finished a one-week trip to China. I've now "survived" all the major (~20) L2 self-driving and robotaxi vehicles in both the US and China. Some thoughts & observations: ▶️L2 self-driving I tested major brands like $Huawei, $Li, $NIO, $Xpeng, and $Xiaomi. Overall, they exceeded my expectations. The rides were not overly cautious and handled complex situations (yes, road conditions in China are very challenging!) quite well. Nothing compares to $Tsla's approach. I see imitation learning/end-to-end as the only effective approach for self-driving. While Chinese peers perform well on main roads, they struggle on frontage roads due to reliance on high-precision maps and rule-based methods (e.g. cars stopped in the middle of the road where there was no clear white lining). Chinese EVs' self-driving capabilities are far ahead of those from US and EU brands. I doubt any Chinese players can profit from L2 self-driving, not because it’s not useful, but because it’s hard to differentiate, and price wars dominate the market in China. Chinese consumers and regulators seem much more receptive to self-driving. Even with a 5/10 self-driving capability, cars are practically hands-free(!) Insurance-wise, for L3+ cars, OEMs bear responsibility for incidents, so OEMs avoid labeling cars as L3+. ▶️Robotaxi I tested major brands like $Didi, and $Bidu. I'd rate equal to $Waymo, and it's ahead of other peers. However, the same issue applies here: user experience is nearly perfect (in Yizhuang, Beijing), but expansion is the real question. Chinese robotaxi companies are very sophisticated. While the rest of the world focuses on technology, Chinese peers treat it as a product, considering unit economics, operations, mass production, etc. Interestingly, most companies expressed a preference NOT to operate fleets themselves. They aim to be asset-light and let fleet managers handle operations. Policy Support: China has a very clear approval process, driven by data (autonomous driving distance, fully driverless distance, intervention rate, passenger ratings, etc.). ▶️Chinese EVs In major cities like Beijing or Shanghai, EV adoption (green license plates vs. gas cars with blue license plates) seems to be 40%+. If 40% of cars on the road are EVs, then EV penetration (defined as the % of new car sales) must already be over 50%. In shopping malls, the ground floor is filled with EV showrooms—easily 10+ brands, many of which are unfamiliar Chinese brands. It appears almost too easy to make an electric car, which is a stark contrast to the US. $Xiaomi, for example, can achieve a 10% gross profit margin in its first year of operation, compared to $RIVN's -45%. Additionally, $Xiaomi cars are priced at 30% of $RIVN's price. It's fascinating to see how China transitioned from "couldn't make their own gas cars at all (only JVs)" to "dominating EVs globally." The government deserves credit for setting the direction and executing effectively. China now controls the entire supply chain, with $CATL holding 40% of the global market share. 🔹How did it happen? The success of the industry Incentives were set just right: the government provided incentives early on to make EVs and gas cars have comparable MSRPs, allowing consumers to choose based on functionality. This approach differs from how the IRA offers incentives... Perfectly competitive market: $TSLA was brought in, and competition was welcomed, unlike the US, which has a 100% import tax on Chinese EVs. Strategic regulations: License plate restrictions were used effectively; for example, taxis and minivans are required to be EVs. 🔹The challenges Despite the success, the industry faces challenges with low-margin companies and struggling stocks. The intense competition shows no sign of ending. Well-funded global OEMs and Chinese state-owned car companies continue to subsidize, leading to new EV brands emerging annually. The natural tendency in China is to race to the bottom. I think this ties back to China's history as the "world’s factory," where manufacturers price products at "cost plus" versus the US and developing countries, which price based on "affordability/value creation." 🔹The wow EV feature >Software features that surprised me the most: - Everything in the car can be voice-controlled. Not just simple tasks like playing music; users can adjust the height of the steering wheel and set the temperature easily. - Self-parking, which $Tsla has yet to release to all FSD users, is already a table stake in China (I'd rate the quality as 10/10). >Other fun hardware features: - Mini fridges in the car - Infotainment systems - IoT: remote access the car/home via cellphone - all connected together - Heads-up displays - UV-protected glass roofs: $Xiaomi took $Tsla's design, but the glass roof of the $Xiaomi car is made of double layers with silver, blocking 99.9% of UV and infrared rays...as a result, heat is no longer a problem inside the car

398,691 görüntüleme

Agents: Quick thoughts & questions on how they operate, their potential, and their limitations A Few Observations - ▶️"Book me a hotel" or "pull historical financials" are already (mostly) solved problems!! Agents can do a ton of tasks right now—like parsing public company press releases and navigating capture key info & complete bookings accurately. However, for more complex navigation flows, the tech still needs some work - but I'm very confident it’s essentially a solved or solvable challenge. ▶️Accuracy & Speed - The key metrics and agents should optimize for. ▶️Lower Build & Migration Costs It took me two minutes to build a new website (link: This is great for consumers—more choices, lower switching costs. Companies will increasingly compete on the quality of their products and services. ▶️Agents vs. Automation tools: The more I think about it, the more I realize that most “agents” are really just automation tools—kind of like how most robots🤖 are just machines lol ❓A Few Key Questions—Would Love Your Thoughts! ❔Remote Servers & Logins In many cases, we’ll want agents to act on our behalf (e.g., log in to to cancel an order). How will platforms like respond? Many websites may block remote servers for security. Is there a technical workaround? ❔Agent Generalization Do we need to train agents on each environment separately, or can one solution handle multiple sites and systems? This seems similar to RL/post-training challenges in AI research. Example: It's unclear to me whether $Devin was specifically trained on environment? ❔Frontend vs. Backend infra for agents to run on I had doubts about Anthropic's "Computer Use" feature, which seemed to run on the frontend, basically remotely controlling my computer so I couldn’t use it at the same time. This should deliver the highest accuracy, but it’s questionable how practical it really is. (Ref: It seems def possible for agents to work quietly “in the background” (like Devin) rather than remotely controlling a user’s PC, but how much accuracy are we sacrificing? A few $Devin test cases that got me thinking: 1⃣Pulling $META's MAU and DAU (1Q21–3Q24) into Excel (video attached) Took Devin 11min - it sent me back an Excel with 100% accurate data. This case was pretty tricky because $Meta changed disclosures and stopped reporting MAU/DAU after 4Q23. Devin didn’t hallucinate data for post-4Q23—it simply didn’t provide it! It really shocked me to see $Devin navigating $Meta's investor relations site (I didn't tell it to find the numbers there), opening each quarterly earnings report, and extracting MAU/DAU like a diligent intern. -> This confirms $Devin (and similar agents) can already accurately “read” screens. 2⃣Booking Hotel (video attached) Devin took 5 minutes to book the InterContinental NYC on after asking for my credentials. From $Devin's workspace, I could see it filling in the correct fields and making the right selections—fast and accurate overall. Interestingly, $Devin didn’t supply all the required information on the first attempt and got some error messages, then retried until it succeeded. It’s unclear whether Devin had been specifically trained on interface or simply learned to adapt on the fly. 3⃣Canceling the Booking This part was even more interesting. While booking didn’t require me to log in, canceling did—so $Devin had to access my (likely via a remote server) account using my Gmail credentials. It successfully canceled the reservation. I wonder how websites will handle future “remote” logins. Notably, Google blocked $Devin’s direct attempts to log in to Gmail when I specifically requested it. 4⃣Booking from Official Hotel Sites I asked Devin to book InterContinental NYC and Four Seasons Boston via their official websites. It made progress but encountered technical hiccups when trying to select the check-in/check-out dates. Insights from Scott Wu on Invest Like the Best: 1/ Self-Driving Cars as the First “Real Agents” Driving requires near-perfect accuracy (99.999%), making it much more demanding than digital or coding agents, which can tolerate more errors. Scott compares $Devin to circa 2014—already good enough to save 90% of your effort, but still short of flawless. 2/ Impact on Collaboration Platforms Tools like Slack and GitLab are likely to see major changes as agents begin to interact with and utilize them along with humans. 2025 should be all about agents - both the disruptors and those they disrupt!

Agents: Quick thoughts & questions on how they operate, their potential, and their limitations A Few Observations - ▶️"Book me a hotel" or "pull historical financials" are already (mostly) solved problems!! Agents can do a ton of tasks right now—like parsing public company press releases and navigating capture key info & complete bookings accurately. However, for more complex navigation flows, the tech still needs some work - but I'm very confident it’s essentially a solved or solvable challenge. ▶️Accuracy & Speed - The key metrics and agents should optimize for. ▶️Lower Build & Migration Costs It took me two minutes to build a new website (link: This is great for consumers—more choices, lower switching costs. Companies will increasingly compete on the quality of their products and services. ▶️Agents vs. Automation tools: The more I think about it, the more I realize that most “agents” are really just automation tools—kind of like how most robots🤖 are just machines lol ❓A Few Key Questions—Would Love Your Thoughts! ❔Remote Servers & Logins In many cases, we’ll want agents to act on our behalf (e.g., log in to to cancel an order). How will platforms like respond? Many websites may block remote servers for security. Is there a technical workaround? ❔Agent Generalization Do we need to train agents on each environment separately, or can one solution handle multiple sites and systems? This seems similar to RL/post-training challenges in AI research. Example: It's unclear to me whether $Devin was specifically trained on environment? ❔Frontend vs. Backend infra for agents to run on I had doubts about Anthropic's "Computer Use" feature, which seemed to run on the frontend, basically remotely controlling my computer so I couldn’t use it at the same time. This should deliver the highest accuracy, but it’s questionable how practical it really is. (Ref: It seems def possible for agents to work quietly “in the background” (like Devin) rather than remotely controlling a user’s PC, but how much accuracy are we sacrificing? A few $Devin test cases that got me thinking: 1⃣Pulling $META's MAU and DAU (1Q21–3Q24) into Excel (video attached) Took Devin 11min - it sent me back an Excel with 100% accurate data. This case was pretty tricky because $Meta changed disclosures and stopped reporting MAU/DAU after 4Q23. Devin didn’t hallucinate data for post-4Q23—it simply didn’t provide it! It really shocked me to see $Devin navigating $Meta's investor relations site (I didn't tell it to find the numbers there), opening each quarterly earnings report, and extracting MAU/DAU like a diligent intern. -> This confirms $Devin (and similar agents) can already accurately “read” screens. 2⃣Booking Hotel (video attached) Devin took 5 minutes to book the InterContinental NYC on after asking for my credentials. From $Devin's workspace, I could see it filling in the correct fields and making the right selections—fast and accurate overall. Interestingly, $Devin didn’t supply all the required information on the first attempt and got some error messages, then retried until it succeeded. It’s unclear whether Devin had been specifically trained on interface or simply learned to adapt on the fly. 3⃣Canceling the Booking This part was even more interesting. While booking didn’t require me to log in, canceling did—so $Devin had to access my (likely via a remote server) account using my Gmail credentials. It successfully canceled the reservation. I wonder how websites will handle future “remote” logins. Notably, Google blocked $Devin’s direct attempts to log in to Gmail when I specifically requested it. 4⃣Booking from Official Hotel Sites I asked Devin to book InterContinental NYC and Four Seasons Boston via their official websites. It made progress but encountered technical hiccups when trying to select the check-in/check-out dates. Insights from Scott Wu on Invest Like the Best: 1/ Self-Driving Cars as the First “Real Agents” Driving requires near-perfect accuracy (99.999%), making it much more demanding than digital or coding agents, which can tolerate more errors. Scott compares $Devin to circa 2014—already good enough to save 90% of your effort, but still short of flawless. 2/ Impact on Collaboration Platforms Tools like Slack and GitLab are likely to see major changes as agents begin to interact with and utilize them along with humans. 2025 should be all about agents - both the disruptors and those they disrupt!

48,850 görüntüleme

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

After months with FSD 12, I’ve gone past the “wow, this is magic” phase. The analyst in me scrutinizes every disengagement and edge case. But today, I just had one of those “tears in my eyes” moments. On a mini road trip, FSD 12.5.2.1 flawlessly handled the winding roads that have always been my worst nightmare. Literally no human I know could’ve done better. Perfectly centered, smooth turns, ideal speed. For two hours, the driver was completely relaxed, finished breakfast, and even the most carsick passenger played on their phone—zero disengagements, perfectly smooth. It passed the “coffee cup test”. Achieving AGI in self-driving is one of the toughest challenges out there, but with 3 trillion miles driven annually and 12 hours per week behind the wheel, it's poised to be one of humanity's greatest unlocks. We’re talking trillion-dollar disruptions that no one’s fully grasping yet—reshaping cities, real estate, traffic, and car ownership! Every step forward gets us closer to that future we all deserve. And today, I felt it.

After months with FSD 12, I’ve gone past the “wow, this is magic” phase. The analyst in me scrutinizes every disengagement and edge case. But today, I just had one of those “tears in my eyes” moments. On a mini road trip, FSD 12.5.2.1 flawlessly handled the winding roads that have always been my worst nightmare. Literally no human I know could’ve done better. Perfectly centered, smooth turns, ideal speed. For two hours, the driver was completely relaxed, finished breakfast, and even the most carsick passenger played on their phone—zero disengagements, perfectly smooth. It passed the “coffee cup test”. Achieving AGI in self-driving is one of the toughest challenges out there, but with 3 trillion miles driven annually and 12 hours per week behind the wheel, it's poised to be one of humanity's greatest unlocks. We’re talking trillion-dollar disruptions that no one’s fully grasping yet—reshaping cities, real estate, traffic, and car ownership! Every step forward gets us closer to that future we all deserve. And today, I felt it.

6,582,586 görüntüleme • 1 yıl önce

Just finished a one-week trip to China. I've now "survived" all the major (~20) L2 self-driving and robotaxi vehicles in both the US and China. Some thoughts & observations: ▶️L2 self-driving I tested major brands like $Huawei, $Li, $NIO, $Xpeng, and $Xiaomi. Overall, they exceeded my expectations. The rides were not overly cautious and handled complex situations (yes, road conditions in China are very challenging!) quite well. Nothing compares to $Tsla's approach. I see imitation learning/end-to-end as the only effective approach for self-driving. While Chinese peers perform well on main roads, they struggle on frontage roads due to reliance on high-precision maps and rule-based methods (e.g. cars stopped in the middle of the road where there was no clear white lining). Chinese EVs' self-driving capabilities are far ahead of those from US and EU brands. I doubt any Chinese players can profit from L2 self-driving, not because it’s not useful, but because it’s hard to differentiate, and price wars dominate the market in China. Chinese consumers and regulators seem much more receptive to self-driving. Even with a 5/10 self-driving capability, cars are practically *hands-free(!)* Insurance-wise, for L3+ cars, OEMs bear responsibility for incidents, so OEMs avoid labeling cars as L3+. ▶️Robotaxi I tested major brands like $Didi, and $Bidu. I'd rate equal to $Waymo, and it's ahead of other peers. However, the same issue applies here: user experience is nearly perfect (in Yizhuang, Beijing), but expansion is the real question. Chinese robotaxi companies are very sophisticated. While the rest of the world focuses on technology, Chinese peers treat it as a product, considering unit economics, operations, mass production, etc. Interestingly, most companies expressed a preference NOT to operate fleets themselves. They aim to be asset-light and let fleet managers handle operations. Policy Support: China has a very clear approval process, driven by data (autonomous driving distance, fully driverless distance, intervention rate, passenger ratings, etc.). ▶️Chinese EVs In major cities like Beijing or Shanghai, EV adoption (green license plates vs. gas cars with blue license plates) seems to be 40%+. If 40% of cars on the road are EVs, then EV penetration (defined as the % of new car sales) must already be over 50%. In shopping malls, the ground floor is filled with EV showrooms—easily 10+ brands, many of which are unfamiliar Chinese brands. It appears almost too easy to make an electric car, which is a stark contrast to the US. $Xiaomi, for example, can achieve a 10% gross profit margin in its first year of operation, compared to $RIVN's -45%. Additionally, $Xiaomi cars are priced at 30% of $RIVN's price. It's fascinating to see how China transitioned from "couldn't make their own gas cars at all (only JVs)" to "dominating EVs globally." The government deserves credit for setting the direction and executing effectively. China now controls the entire supply chain, with $CATL holding 40% of the global market share. 🔹How did it happen? The success of the industry Incentives were set just right: the government provided incentives early on to make EVs and gas cars have comparable MSRPs, allowing consumers to choose based on functionality. This approach differs from how the IRA offers incentives... Perfectly competitive market: $TSLA was brought in, and competition was welcomed, unlike the US, which has a 100% import tax on Chinese EVs. Strategic regulations: License plate restrictions were used effectively; for example, taxis and minivans are required to be EVs. 🔹The challenges Despite the success, the industry faces challenges with low-margin companies and struggling stocks. The intense competition shows no sign of ending. Well-funded global OEMs and Chinese state-owned car companies continue to subsidize, leading to new EV brands emerging annually. The natural tendency in China is to race to the bottom. I think this ties back to China's history as the "world’s factory," where manufacturers price products at "cost plus" versus the US and developing countries, which price based on "affordability/value creation." 🔹The wow EV feature >Software features that surprised me the most: - Everything in the car can be voice-controlled. Not just simple tasks like playing music; users can adjust the height of the steering wheel and set the temperature easily. - Self-parking, which $Tsla has yet to release to all FSD users, is already a table stake in China (I'd rate the quality as 10/10). >Other fun hardware features: - Mini fridges in the car - Infotainment systems - IoT: remote access the car/home via cellphone - all connected together - Heads-up displays - UV-protected glass roofs: $Xiaomi took $Tsla's design, but the glass roof of the $Xiaomi car is made of double layers with silver, blocking 99.9% of UV and infrared rays...as a result, heat is no longer a problem inside the car

Just finished a one-week trip to China. I've now "survived" all the major (~20) L2 self-driving and robotaxi vehicles in both the US and China. Some thoughts & observations: ▶️L2 self-driving I tested major brands like $Huawei, $Li, $NIO, $Xpeng, and $Xiaomi. Overall, they exceeded my expectations. The rides were not overly cautious and handled complex situations (yes, road conditions in China are very challenging!) quite well. Nothing compares to $Tsla's approach. I see imitation learning/end-to-end as the only effective approach for self-driving. While Chinese peers perform well on main roads, they struggle on frontage roads due to reliance on high-precision maps and rule-based methods (e.g. cars stopped in the middle of the road where there was no clear white lining). Chinese EVs' self-driving capabilities are far ahead of those from US and EU brands. I doubt any Chinese players can profit from L2 self-driving, not because it’s not useful, but because it’s hard to differentiate, and price wars dominate the market in China. Chinese consumers and regulators seem much more receptive to self-driving. Even with a 5/10 self-driving capability, cars are practically hands-free(!) Insurance-wise, for L3+ cars, OEMs bear responsibility for incidents, so OEMs avoid labeling cars as L3+. ▶️Robotaxi I tested major brands like $Didi, and $Bidu. I'd rate equal to $Waymo, and it's ahead of other peers. However, the same issue applies here: user experience is nearly perfect (in Yizhuang, Beijing), but expansion is the real question. Chinese robotaxi companies are very sophisticated. While the rest of the world focuses on technology, Chinese peers treat it as a product, considering unit economics, operations, mass production, etc. Interestingly, most companies expressed a preference NOT to operate fleets themselves. They aim to be asset-light and let fleet managers handle operations. Policy Support: China has a very clear approval process, driven by data (autonomous driving distance, fully driverless distance, intervention rate, passenger ratings, etc.). ▶️Chinese EVs In major cities like Beijing or Shanghai, EV adoption (green license plates vs. gas cars with blue license plates) seems to be 40%+. If 40% of cars on the road are EVs, then EV penetration (defined as the % of new car sales) must already be over 50%. In shopping malls, the ground floor is filled with EV showrooms—easily 10+ brands, many of which are unfamiliar Chinese brands. It appears almost too easy to make an electric car, which is a stark contrast to the US. $Xiaomi, for example, can achieve a 10% gross profit margin in its first year of operation, compared to $RIVN's -45%. Additionally, $Xiaomi cars are priced at 30% of $RIVN's price. It's fascinating to see how China transitioned from "couldn't make their own gas cars at all (only JVs)" to "dominating EVs globally." The government deserves credit for setting the direction and executing effectively. China now controls the entire supply chain, with $CATL holding 40% of the global market share. 🔹How did it happen? The success of the industry Incentives were set just right: the government provided incentives early on to make EVs and gas cars have comparable MSRPs, allowing consumers to choose based on functionality. This approach differs from how the IRA offers incentives... Perfectly competitive market: $TSLA was brought in, and competition was welcomed, unlike the US, which has a 100% import tax on Chinese EVs. Strategic regulations: License plate restrictions were used effectively; for example, taxis and minivans are required to be EVs. 🔹The challenges Despite the success, the industry faces challenges with low-margin companies and struggling stocks. The intense competition shows no sign of ending. Well-funded global OEMs and Chinese state-owned car companies continue to subsidize, leading to new EV brands emerging annually. The natural tendency in China is to race to the bottom. I think this ties back to China's history as the "world’s factory," where manufacturers price products at "cost plus" versus the US and developing countries, which price based on "affordability/value creation." 🔹The wow EV feature >Software features that surprised me the most: - Everything in the car can be voice-controlled. Not just simple tasks like playing music; users can adjust the height of the steering wheel and set the temperature easily. - Self-parking, which $Tsla has yet to release to all FSD users, is already a table stake in China (I'd rate the quality as 10/10). >Other fun hardware features: - Mini fridges in the car - Infotainment systems - IoT: remote access the car/home via cellphone - all connected together - Heads-up displays - UV-protected glass roofs: $Xiaomi took $Tsla's design, but the glass roof of the $Xiaomi car is made of double layers with silver, blocking 99.9% of UV and infrared rays...as a result, heat is no longer a problem inside the car

398,691 görüntüleme • 2 yıl önce

Agents: Quick thoughts & questions on how they operate, their potential, and their limitations A Few Observations - ▶️"Book me a hotel" or "pull historical financials" are already (mostly) solved problems!! Agents can do a ton of tasks right now—like parsing public company press releases and navigating capture key info & complete bookings accurately. However, for more complex navigation flows, the tech still needs some work - but I'm very confident it’s essentially a solved or solvable challenge. ▶️Accuracy & Speed - The key metrics and agents should optimize for. ▶️Lower Build & Migration Costs It took me two minutes to build a new website (link: This is great for consumers—more choices, lower switching costs. Companies will increasingly compete on the quality of their products and services. ▶️Agents vs. Automation tools: The more I think about it, the more I realize that most “agents” are really just automation tools—kind of like how most robots🤖 are just machines lol ❓A Few Key Questions—Would Love Your Thoughts! ❔Remote Servers & Logins In many cases, we’ll want agents to act on our behalf (e.g., log in to to cancel an order). How will platforms like respond? Many websites may block remote servers for security. Is there a technical workaround? ❔Agent Generalization Do we need to train agents on each environment separately, or can one solution handle multiple sites and systems? This seems similar to RL/post-training challenges in AI research. Example: It's unclear to me whether $Devin was specifically trained on environment? ❔Frontend vs. Backend infra for agents to run on I had doubts about Anthropic's "Computer Use" feature, which seemed to run on the frontend, basically remotely controlling my computer so I couldn’t use it at the same time. This should deliver the highest accuracy, but it’s questionable how practical it really is. (Ref: It seems def possible for agents to work quietly “in the background” (like Devin) rather than remotely controlling a user’s PC, but how much accuracy are we sacrificing? A few $Devin test cases that got me thinking: 1⃣Pulling $META's MAU and DAU (1Q21–3Q24) into Excel (video attached) Took Devin 11min - it sent me back an Excel with 100% accurate data. This case was pretty tricky because $Meta changed disclosures and stopped reporting MAU/DAU after 4Q23. Devin didn’t hallucinate data for post-4Q23—it simply didn’t provide it! It really shocked me to see $Devin navigating $Meta's investor relations site (I didn't tell it to find the numbers there), opening each quarterly earnings report, and extracting MAU/DAU like a diligent intern. -> This confirms $Devin (and similar agents) can already accurately “read” screens. 2⃣Booking Hotel (video attached) Devin took 5 minutes to book the InterContinental NYC on after asking for my credentials. From $Devin's workspace, I could see it filling in the correct fields and making the right selections—fast and accurate overall. Interestingly, $Devin didn’t supply all the required information on the first attempt and got some error messages, then retried until it succeeded. It’s unclear whether Devin had been specifically trained on interface or simply learned to adapt on the fly. 3⃣Canceling the Booking This part was even more interesting. While booking didn’t require me to log in, canceling did—so $Devin had to access my (likely via a remote server) account using my Gmail credentials. It successfully canceled the reservation. I wonder how websites will handle future “remote” logins. Notably, Google blocked $Devin’s direct attempts to log in to Gmail when I specifically requested it. 4⃣Booking from Official Hotel Sites I asked Devin to book InterContinental NYC and Four Seasons Boston via their official websites. It made progress but encountered technical hiccups when trying to select the check-in/check-out dates. Insights from Scott Wu on Invest Like the Best: 1/ Self-Driving Cars as the First “Real Agents” Driving requires near-perfect accuracy (99.999%), making it much more demanding than digital or coding agents, which can tolerate more errors. Scott compares $Devin to circa 2014—already good enough to save 90% of your effort, but still short of flawless. 2/ Impact on Collaboration Platforms Tools like Slack and GitLab are likely to see major changes as agents begin to interact with and utilize them along with humans. 2025 should be all about agents - both the disruptors and those they disrupt!

Agents: Quick thoughts & questions on how they operate, their potential, and their limitations A Few Observations - ▶️"Book me a hotel" or "pull historical financials" are already (mostly) solved problems!! Agents can do a ton of tasks right now—like parsing public company press releases and navigating capture key info & complete bookings accurately. However, for more complex navigation flows, the tech still needs some work - but I'm very confident it’s essentially a solved or solvable challenge. ▶️Accuracy & Speed - The key metrics and agents should optimize for. ▶️Lower Build & Migration Costs It took me two minutes to build a new website (link: This is great for consumers—more choices, lower switching costs. Companies will increasingly compete on the quality of their products and services. ▶️Agents vs. Automation tools: The more I think about it, the more I realize that most “agents” are really just automation tools—kind of like how most robots🤖 are just machines lol ❓A Few Key Questions—Would Love Your Thoughts! ❔Remote Servers & Logins In many cases, we’ll want agents to act on our behalf (e.g., log in to to cancel an order). How will platforms like respond? Many websites may block remote servers for security. Is there a technical workaround? ❔Agent Generalization Do we need to train agents on each environment separately, or can one solution handle multiple sites and systems? This seems similar to RL/post-training challenges in AI research. Example: It's unclear to me whether $Devin was specifically trained on environment? ❔Frontend vs. Backend infra for agents to run on I had doubts about Anthropic's "Computer Use" feature, which seemed to run on the frontend, basically remotely controlling my computer so I couldn’t use it at the same time. This should deliver the highest accuracy, but it’s questionable how practical it really is. (Ref: It seems def possible for agents to work quietly “in the background” (like Devin) rather than remotely controlling a user’s PC, but how much accuracy are we sacrificing? A few $Devin test cases that got me thinking: 1⃣Pulling $META's MAU and DAU (1Q21–3Q24) into Excel (video attached) Took Devin 11min - it sent me back an Excel with 100% accurate data. This case was pretty tricky because $Meta changed disclosures and stopped reporting MAU/DAU after 4Q23. Devin didn’t hallucinate data for post-4Q23—it simply didn’t provide it! It really shocked me to see $Devin navigating $Meta's investor relations site (I didn't tell it to find the numbers there), opening each quarterly earnings report, and extracting MAU/DAU like a diligent intern. -> This confirms $Devin (and similar agents) can already accurately “read” screens. 2⃣Booking Hotel (video attached) Devin took 5 minutes to book the InterContinental NYC on after asking for my credentials. From $Devin's workspace, I could see it filling in the correct fields and making the right selections—fast and accurate overall. Interestingly, $Devin didn’t supply all the required information on the first attempt and got some error messages, then retried until it succeeded. It’s unclear whether Devin had been specifically trained on interface or simply learned to adapt on the fly. 3⃣Canceling the Booking This part was even more interesting. While booking didn’t require me to log in, canceling did—so $Devin had to access my (likely via a remote server) account using my Gmail credentials. It successfully canceled the reservation. I wonder how websites will handle future “remote” logins. Notably, Google blocked $Devin’s direct attempts to log in to Gmail when I specifically requested it. 4⃣Booking from Official Hotel Sites I asked Devin to book InterContinental NYC and Four Seasons Boston via their official websites. It made progress but encountered technical hiccups when trying to select the check-in/check-out dates. Insights from Scott Wu on Invest Like the Best: 1/ Self-Driving Cars as the First “Real Agents” Driving requires near-perfect accuracy (99.999%), making it much more demanding than digital or coding agents, which can tolerate more errors. Scott compares $Devin to circa 2014—already good enough to save 90% of your effort, but still short of flawless. 2/ Impact on Collaboration Platforms Tools like Slack and GitLab are likely to see major changes as agents begin to interact with and utilize them along with humans. 2025 should be all about agents - both the disruptors and those they disrupt!

48,850 görüntüleme • 1 yıl önce

Daha fazla içerik yok.