TechCrunch Summary (2023-05-10 10AM)

Source: Yeah, tech growth is slowing down

Summary
Airbnb, Amplitude, and Twilio suffered in the stock market after disclosing lower than expected growth forecasts for the year. Airbnb’s age in the industry and GAAP profit in Q1 couldn’t help the company overcome its predicted growth of 12% to 16% in Q2 2022. Twilio also posted better than expected financials for Q1, but analysts had hoped for better revenue than the company’s forecast of just 4% to 5% growth for Q2.

Fact

📉 Airbnb’s revenue is expected to increase only by 12% to 16% in Q2 2022, which is less than what the company witnessed in Q2 2020 (58%) and even less than its Q1 2023 growth (20%).

📈 Twilio reported better-than-expected profit and revenue for Q1, but market analysts were expecting a far greater $1.05 billion in revenue for Q2, while Twilio’s forecast stands at $980 million to $990 million, which is just a 4% to 5% increase.

📊 The smaller peers of big tech companies are also seeing headwinds in growth.

---

Source: Google Maps is getting a new ‘Immersive View for Routes’ feature

Summary
Google Maps is launching “Immersive View for Routes”, a new feature that provides multidimensional previews of a user’s journey, including information about traffic, bike lanes, parking, air quality, intersections, and more. The feature will be rolling out over the next few months in 15 cities worldwide, including London, Los Angeles, New York, and Tokyo.

Facts

🌆 Immersive View for Routes provides multidimensional previews of a user’s journey.

🗺️ The feature will launch in 15 cities, including London, Los Angeles, and New York.

🚲 Users can preview bike lanes, sidewalks, intersections, and parking along their journey.

☀️ Immersive View for Routes lets users see information about air quality and weather for their journey.

🎥 Google has also launched an Aerial View API and Photorealistic 3D titles for developers.

---

Source: Google’s PaLM 2 paper shows that text-generating AI still has a long way to go

Summary
Google claims that its PaLM 2 multilingual text model is a significant improvement over its predecessor and even surpasses OpenAI’s GPT-4 in some aspects, although limited detail about its technical features is available. PaLM 2 carried out well on accuracy and quality criteria but was less effective in certain tests, displaying bias and producing toxic responses, especially for prompts containing implicit or subtle harm.

Fact

🔍 PaLM 2 is claimed to be a significant improvement over PaLM for understanding and generating multilingual text.

❓ The technical specifications of PaLM 2 are not clear due to Google’s recent policy of keeping strategic AI research in-house to compete with Microsoft and OpenAI.

🔍 Google reveals that the parameters of the most capable PaLM 2 model are 14.7 billion.

❓ PaLM 2’s training data, which includes a higher percentage of non-English data, has not been disclosed.

😕 PaLM 2 generated toxic responses over 30% of the time in response to explicitly toxic stimuli and was even more toxic (60%) in response to implicitly harmful stimuli.

😒 The model gave a toxic response almost 18% of the time in one bias test, especially regarding English, German, and Portuguese prompts related to racial issues.

---

Source: Google launches PaLM 2, its next-gen large language model

Summary
Google announced the launch of PaLM 2, its latest large language model which can solve math puzzles, perform common sense reasoning, and supports writing and debugging code. It was trained on 20 programming languages and can process 20 tokens per second on smartphones. It also has improved multilingual support and has a smaller version. PaLM 2 is available through Google’s PaLM API, Firebase and Colab.

Facts

💬 PaLM 2 is Google’s latest large language model.

🤖 PaLM 2 can solve math puzzles and perform common sense reasoning.

💻 It also features improved support for writing and debugging code and was trained on 20 programming languages.

🌍 PaLM 2 was trained on a corpus that features over 100 languages and has improved multilingual support.

📱 A smaller version of PaLM 2 can run on smartphones, processing 20 tokens per second.

---

Source: Google Photos to gain a new ‘Magic Editor’ feature powered by generative AI

Summary
Google Photos introduced Magic Editor, a new tool that leverages AI techniques, including generative AI for editing images. The feature will enable users to remove specific parts of photos and replace them with AI-generated new elements. The tool will also allow the repositioning of image subjects and filling gaps in photos. Magic Editor will be released as an experimental feature later this year.

Fact
💡 Google Photos will introduce Magic Editor, a new tool that uses AI techniques for editing images.
💡 Magic Editor will allow users to remove specific parts of the photo, replace them with AI-generated new elements, reposition image subjects, and fill gaps in photos.
💡 The tool will be made available as an experimental feature.
💡 It is unclear if Google will charge for this feature.

---